Links to Exploration step
Le document en format XML
<record><TEI><teiHeader><fileDesc><titleStmt><title xml:lang="en">Connecting Neurons to a Mobile Robot: An In Vitro Bidirectional Neural Interface</title>
<author><name sortKey="Novellino, A" sort="Novellino, A" uniqKey="Novellino A" first="A." last="Novellino">A. Novellino</name>
<affiliation><nlm:aff id="I1">Neuroengineering and Bio-nanotechnology Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genova, Via Opera Pia 11a, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="D Angelo, P" sort="D Angelo, P" uniqKey="D Angelo P" first="P." last="D'Angelo">P. D'Angelo</name>
<affiliation><nlm:aff id="I2">NeuroLab, Department of Informatics Systems and Telematics (DIST), Via Opera Pia 13, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Cozzi, L" sort="Cozzi, L" uniqKey="Cozzi L" first="L." last="Cozzi">L. Cozzi</name>
<affiliation><nlm:aff id="I2">NeuroLab, Department of Informatics Systems and Telematics (DIST), Via Opera Pia 13, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Chiappalone, M" sort="Chiappalone, M" uniqKey="Chiappalone M" first="M." last="Chiappalone">M. Chiappalone</name>
<affiliation><nlm:aff id="I1">Neuroengineering and Bio-nanotechnology Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genova, Via Opera Pia 11a, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Sanguineti, V" sort="Sanguineti, V" uniqKey="Sanguineti V" first="V." last="Sanguineti">V. Sanguineti</name>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Martinoia, S" sort="Martinoia, S" uniqKey="Martinoia S" first="S." last="Martinoia">S. Martinoia</name>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
</author>
</titleStmt>
<publicationStmt><idno type="wicri:source">PMC</idno>
<idno type="pmid">18350128</idno>
<idno type="pmc">2266971</idno>
<idno type="url">http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2266971</idno>
<idno type="RBID">PMC:2266971</idno>
<idno type="doi">10.1155/2007/12725</idno>
<date when="2007">2007</date>
<idno type="wicri:Area/Pmc/Corpus">000442</idno>
<idno type="wicri:explorRef" wicri:stream="Pmc" wicri:step="Corpus" wicri:corpus="PMC">000442</idno>
</publicationStmt>
<sourceDesc><biblStruct><analytic><title xml:lang="en" level="a" type="main">Connecting Neurons to a Mobile Robot: An In Vitro Bidirectional Neural Interface</title>
<author><name sortKey="Novellino, A" sort="Novellino, A" uniqKey="Novellino A" first="A." last="Novellino">A. Novellino</name>
<affiliation><nlm:aff id="I1">Neuroengineering and Bio-nanotechnology Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genova, Via Opera Pia 11a, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="D Angelo, P" sort="D Angelo, P" uniqKey="D Angelo P" first="P." last="D'Angelo">P. D'Angelo</name>
<affiliation><nlm:aff id="I2">NeuroLab, Department of Informatics Systems and Telematics (DIST), Via Opera Pia 13, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Cozzi, L" sort="Cozzi, L" uniqKey="Cozzi L" first="L." last="Cozzi">L. Cozzi</name>
<affiliation><nlm:aff id="I2">NeuroLab, Department of Informatics Systems and Telematics (DIST), Via Opera Pia 13, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Chiappalone, M" sort="Chiappalone, M" uniqKey="Chiappalone M" first="M." last="Chiappalone">M. Chiappalone</name>
<affiliation><nlm:aff id="I1">Neuroengineering and Bio-nanotechnology Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genova, Via Opera Pia 11a, 16145 Genova, Italy</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Sanguineti, V" sort="Sanguineti, V" uniqKey="Sanguineti V" first="V." last="Sanguineti">V. Sanguineti</name>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
</author>
<author><name sortKey="Martinoia, S" sort="Martinoia, S" uniqKey="Martinoia S" first="S." last="Martinoia">S. Martinoia</name>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
<affiliation><nlm:aff>NONE</nlm:aff>
</affiliation>
</author>
</analytic>
<series><title level="j">Computational Intelligence and Neuroscience</title>
<idno type="ISSN">1687-5265</idno>
<idno type="eISSN">1687-5273</idno>
<imprint><date when="2007">2007</date>
</imprint>
</series>
</biblStruct>
</sourceDesc>
</fileDesc>
<profileDesc><textClass></textClass>
</profileDesc>
</teiHeader>
<front><div type="abstract" xml:lang="en"><p>One of the key properties of intelligent behaviors is the capability to learn and adapt to changing environmental conditions. These features are the result of the continuous and intense interaction of the brain with the external world, mediated by the body. For this reason “embodiment” represents an innovative and very suitable experimental paradigm when studying the neural processes underlying learning new behaviors and adapting to unpredicted situations. To this purpose, we developed a novel bidirectional neural interface. We interconnected in vitro neurons, extracted from rat embryos and plated on a microelectrode array (MEA), to external devices, thus allowing real-time closed-loop interaction. The novelty of this experimental approach entails the necessity to explore different computational schemes and experimental hypotheses. In this paper, we present an open, scalable architecture, which allows fast prototyping of different modules and where coding and decoding schemes and different experimental configurations can be tested. This hybrid system can be used for studying the computational properties and information coding in biological neuronal networks with far-reaching implications for the future development of advanced neuroprostheses.</p>
</div>
</front>
<back><div1 type="bibliography"><listBibl><biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
<biblStruct></biblStruct>
</listBibl>
</div1>
</back>
</TEI>
<pmc article-type="research-article"><pmc-dir>properties open_access</pmc-dir>
<front><journal-meta><journal-id journal-id-type="nlm-ta">Comput Intell Neurosci</journal-id>
<journal-id journal-id-type="publisher-id">CIN</journal-id>
<journal-title>Computational Intelligence and Neuroscience</journal-title>
<issn pub-type="ppub">1687-5265</issn>
<issn pub-type="epub">1687-5273</issn>
<publisher><publisher-name>Hindawi Publishing Corporation</publisher-name>
</publisher>
</journal-meta>
<article-meta><article-id pub-id-type="pmid">18350128</article-id>
<article-id pub-id-type="pmc">2266971</article-id>
<article-id pub-id-type="doi">10.1155/2007/12725</article-id>
<article-categories><subj-group subj-group-type="heading"><subject>Research Article</subject>
</subj-group>
</article-categories>
<title-group><article-title>Connecting Neurons to a Mobile Robot: An In Vitro Bidirectional Neural Interface</article-title>
</title-group>
<contrib-group><contrib contrib-type="author"><name><surname>Novellino</surname>
<given-names>A.</given-names>
</name>
<xref ref-type="aff" rid="I1"><sup>1</sup>
</xref>
<xref ref-type="corresp" rid="cor1">*</xref>
</contrib>
<contrib contrib-type="author"><name><surname>D'Angelo</surname>
<given-names>P.</given-names>
</name>
<xref ref-type="aff" rid="I2"><sup>2</sup>
</xref>
</contrib>
<contrib contrib-type="author"><name><surname>Cozzi</surname>
<given-names>L.</given-names>
</name>
<xref ref-type="aff" rid="I2"><sup>2</sup>
</xref>
</contrib>
<contrib contrib-type="author"><name><surname>Chiappalone</surname>
<given-names>M.</given-names>
</name>
<xref ref-type="aff" rid="I1"><sup>1</sup>
</xref>
</contrib>
<contrib contrib-type="author"><name><surname>Sanguineti</surname>
<given-names>V.</given-names>
</name>
<xref ref-type="aff" rid="I2"><sup>2, 3</sup>
</xref>
<xref ref-type="aff" rid="I3"></xref>
</contrib>
<contrib contrib-type="author"><name><surname>Martinoia</surname>
<given-names>S.</given-names>
</name>
<xref ref-type="aff" rid="I1"><sup>1, 3</sup>
</xref>
<xref ref-type="aff" rid="I3"></xref>
</contrib>
</contrib-group>
<aff id="I1"><sup>1</sup>
Neuroengineering and Bio-nanotechnology Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genova, Via Opera Pia 11a, 16145 Genova, Italy</aff>
<aff id="I2"><sup>2</sup>
NeuroLab, Department of Informatics Systems and Telematics (DIST), Via Opera Pia 13, 16145 Genova, Italy</aff>
<aff id="I3"><sup>3</sup>
Center for Neuroscience and Neuroengineering “Massimo Grattarola”, University of Genova, 16132 Genova,Viale Benedetto XV, 3, Italy</aff>
<author-notes><corresp id="cor1">*Antonio Novellino: <email>antonio.novellino@ettsolutions.com</email>
</corresp>
<fn fn-type="other"><p>Recommended by Fabio Babiloni</p>
</fn>
</author-notes>
<pub-date pub-type="ppub"><year>2007</year>
</pub-date>
<pub-date pub-type="epub"><day>30</day>
<month>7</month>
<year>2007</year>
</pub-date>
<volume>2007</volume>
<elocation-id>12725</elocation-id>
<history><date date-type="received"><day>27</day>
<month>12</month>
<year>2006</year>
</date>
<date date-type="rev-recd"><day>4</day>
<month>4</month>
<year>2007</year>
</date>
<date date-type="accepted"><day>18</day>
<month>6</month>
<year>2007</year>
</date>
</history>
<permissions><copyright-statement>Copyright © 2007 A. Novellino et al.</copyright-statement>
<copyright-year>2007</copyright-year>
<license license-type="open-access"><p>This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>
</license>
</permissions>
<abstract><p>One of the key properties of intelligent behaviors is the capability to learn and adapt to changing environmental conditions. These features are the result of the continuous and intense interaction of the brain with the external world, mediated by the body. For this reason “embodiment” represents an innovative and very suitable experimental paradigm when studying the neural processes underlying learning new behaviors and adapting to unpredicted situations. To this purpose, we developed a novel bidirectional neural interface. We interconnected in vitro neurons, extracted from rat embryos and plated on a microelectrode array (MEA), to external devices, thus allowing real-time closed-loop interaction. The novelty of this experimental approach entails the necessity to explore different computational schemes and experimental hypotheses. In this paper, we present an open, scalable architecture, which allows fast prototyping of different modules and where coding and decoding schemes and different experimental configurations can be tested. This hybrid system can be used for studying the computational properties and information coding in biological neuronal networks with far-reaching implications for the future development of advanced neuroprostheses.</p>
</abstract>
</article-meta>
</front>
<body><sec sec-type="section" id="sec1"><title>1. INTRODUCTION</title>
<p>Electrophysiological techniques, both in vivo and in
vitro, are traditionally used to study spontaneous neural activity and its
modifications in response to different kinds of external stimuli (e.g.,
chemical, electrical, electromagnetic). One of the main limitations of these
studies is the total absence of a sensory and motor “context.” This condition
is particularly unnatural: complex mechanisms, like learning, are the result of
a continuous interaction between the nervous system and the environment,
mediated by the body. For this reason, during the last years, the growing
interest in neuroscience for closed-loop experiments (cf. Society for
Neuroscience Meeting 2004, San Diego (Calif, USA); <ext-link ext-link-type="uri" xlink:href="http://apu.sfn.org">http://apu.sfn.org</ext-link>
) has led
to the development of several innovative bidirectional platforms, under the
hypothesis that the dynamical and adaptive properties of neural systems may be
better understood in the context of the interaction between the brain and the
external environment.</p>
<p>In the last few years, interaction has been studied at
different levels of investigation: at the molecular level, by synthesizing the
behavior of artificial ion channels—the dynamic-clamp technique
(Sharp et al. [<xref ref-type="bibr" rid="B42">2</xref>
]); at the single neuron
level, by interfacing artificial and actual neurons (Le Masson et al. [<xref ref-type="bibr" rid="B21">3</xref>
]); at the population level, by controlling the dynamic
regime of neuronal populations (Wagenaar et al. [<xref ref-type="bibr" rid="B20">4</xref>
]) and its
adaptive properties (Shahaf and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]; Marom and Eytan [<xref ref-type="bibr" rid="B24">6</xref>
]); and, finally, at the whole system level, by means of
experiments in which portions of the ex vivo/in vivo brain of an animal are
connected to artificial/virtual robots to form bioartificial/hybrid systems
(Reger et al. [<xref ref-type="bibr" rid="B44">7</xref>
]; Wessberg et al. [<xref ref-type="bibr" rid="B34">8</xref>
]; Nicolelis [<xref ref-type="bibr" rid="B22">9</xref>
]; Schwartz et al. [<xref ref-type="bibr" rid="B18">10</xref>
];
Karniel et al. [<xref ref-type="bibr" rid="B46">11</xref>
]).</p>
<p>An alternative and simplified paradigm to study the
interaction between the brain and the external world is the “embodied electrophysiology,”
where dissociated neuronal networks are bidirectionally coupled to artificial
systems (DeMarse et al. [<xref ref-type="bibr" rid="B47">12</xref>
]; Bakkum et al. [<xref ref-type="bibr" rid="B12">13</xref>
]; Martinoia et al. [<xref ref-type="bibr" rid="B37">14</xref>
]), which
provide a physical body to the in vitro brain and allow it to interact with the
environment (Potter [<xref ref-type="bibr" rid="B27">15</xref>
]). This paradigm can be used to investigate the
mechanisms that the nervous system uses to represent, store, and process sensory-motor
information, understanding how living neurons lead to higher-level cognition
and “intelligent behavior” (Bakkum et al. [<xref ref-type="bibr" rid="B12">13</xref>
]).</p>
<p>The development of in vitro bidirectional neural
interface offers the unique opportunity to explore the adaptive properties of a
model of the neural system and it can be of valuable help for the future
developments of in vivo neural interfaces (Mussa-Ivaldi and Miller [<xref ref-type="bibr" rid="B29">16</xref>
]; Nicolelis [<xref ref-type="bibr" rid="B23">17</xref>
]). Ideally, in vivo
brain-machine interfaces should enable two-way communication, that is, both
stimulation and recording at the same time. Two-way interaction would be
particularly crucial in advanced neuroprostheses. Sensory systems cannot be
fully restored by simply mapping input into the brain; instead, neuroprosthetic
devices should be fused with the reciprocating neural interactivity that is
responsible for ongoing conscious awareness.</p>
<p>The aim of this paper is to describe the architecture
and the high potential of the developed neurorobotic system, that is, a
neuronal network connected to a mobile robot. In the “methods” section, we
discuss the issues underlying design and computational choices. The
computational requirements for the closed-loop system are very demanding,
mainly due to the necessity to simultaneously process high-frequency
multichannel data, in real time. On the other hand, the novelty of this
approach involves the necessity to explore different computational schemes
(e.g., to change the coding/decoding strategy, the number of input/output
electrodes, and the value of the parameters). In the “results” section, we
describe the computational performances of the developed system and the
strategies for selecting the input and output sites, an essential step when
dealing with neuronal model with a no predefined architecture, such as
dissociated cultures (see Figures <xref ref-type="fig" rid="fig1">Figure1(a)</xref>
, <xref ref-type="fig" rid="fig1">Figure1(b)</xref>
). Finally, the preliminary
experiments involving a network of cortical neurons and a robotic body are
presented and the main improvements with respect to our previous results
(Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]) are underlined, both in
terms of computational architecture and experimental protocol. The use of a
simple reactive behavior (i.e., obstacle avoidance) demonstrates the
feasibility of the approach and the potential of this novel experimental
paradigm.</p>
</sec>
<sec sec-type="section" id="sec2"><title>2. MATERIALS AND METHODS</title>
<sec sec-type="subsection" id="subsec2.1"><title>2.1. Robot body, playground arena and obstacle avoidance task</title>
<p>Modeling of adaptive behavior by developing adaptive
autonomous agents is an approach widely investigated in the fields of
artificial intelligence and autonomous control (Maes [<xref ref-type="bibr" rid="B40">19</xref>
]; Brooks [<xref ref-type="bibr" rid="B31">20</xref>
]), and a particular model of
adaptive behavior is represented by an agent who is motivated in trying to
survive in a defined environment, without any external (i.e., human) help.</p>
<p>The agent may generate its actions exclusively from
the available sensory information, or may use some kind of previous
“experience.” The former type of agent is generally referred as “reactive.”
One of the most studied implementations of this model is the “exploring”
vehicle paradigm. In 1984, Braitenberg [<xref ref-type="bibr" rid="B11">21</xref>
] proposed a simple architecture, that is, a vehicle
with direct links between sensors and motors (the greater the sensors values
are, the faster the motors run), that seems to mimic an intelligent behavior in
a real context. The easiest example of the Braitenberg's vehicles is a
two-wheeled robot with two light sensors that, according to the connection
between sensors and motor-wheels, can produce different and interesting
behaviors (fear, aggressiveness). Here we will show a neurorobotic Braitenberg
“explorer” vehicle as an example of application of the closed-loop platform
for embodied electrophysiology.</p>
<p>The artificial body consists of a small mobile robot
(Khepera II, K-team, <ext-link ext-link-type="uri" xlink:href="http://www.k-team.com">http://www.k-team.com</ext-link>
), equipped with two wheels and eight
infrared (IR) proximity sensors that provide information about the distance of
the robot from obstacles. In our experiments, the robot (7 cm diameter) moves
inside a circular arena (80 cm diameter), containing wooden cylindrical
obstacles (7 cm diameter). The Khepera robot and its playground are shown in
<xref ref-type="fig" rid="fig1">Figure 1(c)</xref>
. In order to partially compensate the high nonlinearity of the
proximity sensors and the influence of the ambient light, the internal
perimeter of the playground and the border of the obstacles were covered with
an IR reflective tape.</p>
</sec>
<sec sec-type="subsection" id="subsec2.2"><title>2.2. Computational architecture of the neural interface</title>
<p>To establish a bidirectional communication between the
neuronal preparation and a mobile robot, the electrophysiological signals need
to be translated into motor commands for the robot (decoding of neural
activity), and at the same time the sensory signal from the robot need to be
translated into a pattern of electrical stimulation (coding of sensory
information). <xref ref-type="fig" rid="fig2">Figure 2</xref>
presents the general computational architecture of the
proposed closed-loop system that can be summarized in the following three main
parts (i.e., from left to right in <xref ref-type="fig" rid="fig2">Figure 2</xref>
).</p>
<p><list list-type="order"><list-item><p>Coding (from
the robot to the neural preparation): while the robot freely moves into the
playground, its IR sensors see whether or not an obstacle is in the proximity
and where it is (left or right side). The IR signals <italic>u</italic>
(<italic>t</italic>
) are weighted
according to the sensory receptive field law and the two resulting stimulation
signals <italic>s</italic>
(<italic>t</italic>
), relative to the right and left “eye” of the robot,
are then coded into a feedback stimulation <italic>x</italic>
(<italic>t</italic>
)</p>
</list-item>
<list-item><p>Processing of
electrophysiological signals: the spontaneous or evoked electrophysiological
activity <italic>y</italic>
(<italic>t</italic>
) is sampled <inline-formula><mml:math id="M1"><mml:mrow><mml:mo stretchy="false">(</mml:mo>
<mml:mover accent="true"><mml:mi>y</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
and processed,
in order to give an estimation of the instantaneous firing rate <inline-formula><mml:math id="M2"><mml:mrow><mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>y</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
</p>
</list-item>
<list-item><p>Decoding (from
the neural preparation to the robot): the processed electrophysiological signal <inline-formula><mml:math id="M3"><mml:mrow><mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>y</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
is translated
into motor commands <italic>ω</italic>
(<italic>t</italic>
for the robot,
according to the specified decoding strategy.</p>
</list-item>
</list>
</p>
<p>To make our
computational architecture as open as possible, we developed a library of
coding/decoding schemes (Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]; Cozzi et al. [<xref ref-type="bibr" rid="B3">22</xref>
]), and identified the most effective ones in achieving
a desired behavioral task.</p>
<statement id="head1"><title>Library of coding schemes</title>
<p>Coding means the representation of external sensory
input patterns in terms of electrical stimulation and hence of neuron's
activity. In this perspective, the implemented <italic>neural code</italic>
has been of
two main typologies, both of them based on the rate coding concept.</p>
<p><list list-type="order"><list-item><p>Proportional coding.</p>
<p>The rate of stimulation <italic>r</italic>
<sub>s</sub>
(<italic>t</italic>
) is proportional
to the sensory feedback <italic>s</italic>
(<italic>t</italic>
). The maximum rate of stimulation, <italic>r</italic>
<sub><italic>s</italic>
</sub>
<sup><italic>max</italic>
</sup>
, is only attained when the robot hits an obstacle. It
was suggested that this value has to be as large as possible for accurate
coding of the temporal structure of sensory signal (Wagenaar et al. [<xref ref-type="bibr" rid="B13">1</xref>
], DeMarse et al. [<xref ref-type="bibr" rid="B47">12</xref>
]), but at
the same time it has to be low in order not to damage the culture (Shahaf and
Marom [<xref ref-type="bibr" rid="B25">5</xref>
]), therefore,
the maximum of the stimulation rate <italic>r</italic>
<italic><sub>s</sub>
</italic>
(<italic>t</italic>
) was up to 2 Hz.</p>
</list-item>
<list-item><p>Binary coding.</p>
<p>A binary coding scheme generates trigger signals for
the electrical stimulator only when the sensory feedback <italic>s</italic>
(<italic>t</italic>
) overcomes a
threshold, approximately reflecting the presence of an obstacle at 5 cm
distance. The stimulation rate <italic>r</italic>
<sub><italic>s</italic>
</sub>
(<italic>t</italic>
) is therefore
either 0 or 1 Hz. The maximum frequency of stimulation was chosen according to
what reported in literature (Shahaf and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]).</p>
</list-item>
</list>
</p>
</statement>
<statement id="head2"><title>Preprocessing of electrophysiological signal</title>
<p> <italic>Spike detection</italic>
. The electrophysiological signals <inline-formula><mml:math id="M4"><mml:mrow><mml:mover accent="true"><mml:mi>y</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
acquired from
MEA electrodes must be preprocessed in order to remove the stimulus artifact
and to isolate spikes from noise.</p>
<p>The spike detection algorithm uses a differential
peak-to-peak threshold to follow the variability of the signal (Grattarola et al. [<xref ref-type="bibr" rid="B28">23</xref>
]). A time window, sized to contain at most one single
spike (4 ms), is shifted along the signal, sampled at the frequency of 10 kHz.
Within the window when the difference between the maximum and the minimum
exceeds the threshold, a spike is found and its time stamp is saved. In this
way, the resulting spike train signal is sampled at 250 Hz. The threshold is
proportional to the noise standard deviation (SD) and is calculated separately
for each individual channel (typically as 7 × SD) before the
beginning of the actual experiment (i.e., the spontaneous activity recording,
see Section 3.3).</p>
<p> <italic>Blanking of stimulus artifact</italic>
. Stimulus artifacts are detected when the recorded
signal exceeds a second, higher threshold. The artifact is then suppressed by
cancelling the first sample in the spike train occurring immediately after it,
corresponding to a signal blanking of 4 milliseconds after stimulus delivery.
This quite conservative procedure could have been improved, but we found it
effective for our applications.</p>
</statement>
<statement id="head3"><title>Decoding schemes</title>
<p>Although several linear and nonlinear algorithms for
translating neuronal activity into motor commands for external actuators have
been proposed (Chapin et al. [<xref ref-type="bibr" rid="B30">24</xref>
]; Wessberg et al. [<xref ref-type="bibr" rid="B34">8</xref>
];
Carmena et al. [<xref ref-type="bibr" rid="B19">25</xref>
]; Wessberg and Nicolelis
[<xref ref-type="bibr" rid="B32">26</xref>
]), here the
decoding schemes are simply based on rate-coding (Fetz [<xref ref-type="bibr" rid="B6">27</xref>
]), that has proven to be
efficient in brain-machine-interfaces (Lebedev and Nicolelis [<xref ref-type="bibr" rid="B33">28</xref>
]).</p>
<p> <italic>Firing rate estimation.</italic>
The neural activity is represented by the
instantaneous firing rate <inline-formula><mml:math id="M5"><mml:mrow><mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>y</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
on each
recording channel and it is estimated from the spike trains <italic>y</italic>
(<italic>t</italic>
) through a
low-pass filter. Two different filters have been implemented (a first-order and
a second-order filter).</p>
<p> <italic>Decoding.</italic>
The recording sites are divided into two groups, respectively used for
controlling the left and right wheel, each of them formed by <italic>N</italic>
electrodes. The
motor commands <italic>ω</italic>
(<italic>t</italic>
), that is, the angular speeds of the wheels, are
obtained by implementing the following winner-takes-all (WTA)
mechanism:<disp-formula id="eq1"><label>(1)</label>
<mml:math id="M6"><mml:mrow><mml:mtable columnalign="left"><mml:mtr columnalign="left"><mml:mtd columnalign="left"><mml:mrow><mml:msub><mml:mi>ω</mml:mi>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:mtd>
<mml:mtd columnalign="left"><mml:mrow><mml:mo>=</mml:mo>
<mml:mrow><mml:mo stretchy="true">{</mml:mo>
<mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mo stretchy="true">(</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mn>0</mml:mn>
</mml:msub>
<mml:mo>−</mml:mo>
<mml:mstyle><mml:munderover><mml:mo>∑</mml:mo>
<mml:mrow><mml:mi>i</mml:mi>
<mml:mo>=</mml:mo>
<mml:mn>1</mml:mn>
</mml:mrow>
<mml:mi>N</mml:mi>
</mml:munderover>
</mml:mstyle>
<mml:msub><mml:mi>C</mml:mi>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo>⋅</mml:mo>
<mml:msub><mml:mrow><mml:mo stretchy="false">[</mml:mo>
<mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
<mml:mo stretchy="false">]</mml:mo>
</mml:mrow>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo stretchy="true">)</mml:mo>
</mml:mrow>
</mml:mtd>
<mml:mtd><mml:mrow><mml:mtext>if</mml:mtext>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo>≥</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:mtd>
</mml:mtr>
<mml:mtr><mml:mtd><mml:mrow><mml:mo>−</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>b</mml:mi>
</mml:msub>
</mml:mrow>
</mml:mtd>
<mml:mtd><mml:mrow><mml:mtext>if</mml:mtext>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo><</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:mtd>
</mml:mtr>
</mml:mtable>
</mml:mrow>
</mml:mrow>
</mml:mrow>
</mml:mtd>
</mml:mtr>
<mml:mtr columnalign="left"><mml:mtd columnalign="left"><mml:mrow><mml:msub><mml:mi>ω</mml:mi>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:mtd>
<mml:mtd columnalign="left"><mml:mrow><mml:mo>=</mml:mo>
<mml:mrow><mml:mo stretchy="true">{</mml:mo>
<mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mo stretchy="true">(</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mn>0</mml:mn>
</mml:msub>
<mml:mo>−</mml:mo>
<mml:mstyle><mml:munderover><mml:mo>∑</mml:mo>
<mml:mrow><mml:mi>i</mml:mi>
<mml:mo>=</mml:mo>
<mml:mn>1</mml:mn>
</mml:mrow>
<mml:mi>N</mml:mi>
</mml:munderover>
</mml:mstyle>
<mml:msub><mml:mi>C</mml:mi>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo>⋅</mml:mo>
<mml:msub><mml:mrow><mml:mo stretchy="false">[</mml:mo>
<mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
<mml:mo stretchy="false">]</mml:mo>
</mml:mrow>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo stretchy="true">)</mml:mo>
</mml:mrow>
</mml:mtd>
<mml:mtd><mml:mrow><mml:mtext>if</mml:mtext>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo>≥</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:mtd>
</mml:mtr>
<mml:mtr><mml:mtd><mml:mrow><mml:mo>−</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>b</mml:mi>
</mml:msub>
</mml:mrow>
</mml:mtd>
<mml:mtd><mml:mrow><mml:mtext>if</mml:mtext>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>R</mml:mi>
</mml:msub>
<mml:mo><</mml:mo>
<mml:msub><mml:mi>ω</mml:mi>
<mml:mi>L</mml:mi>
</mml:msub>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:mtd>
</mml:mtr>
</mml:mtable>
</mml:mrow>
</mml:mrow>
</mml:mrow>
</mml:mtd>
</mml:mtr>
</mml:mtable>
</mml:mrow>
</mml:math>
</disp-formula>
where <italic>ω</italic>
<sub><italic>b</italic>
</sub>
is a constant
angular speed (up to 2 rad/s), <italic>ω</italic>
<sub><italic>0</italic>
</sub>
is the maximum
angular speed (i.e., 5 rad/s); <inline-formula><mml:math id="M7"><mml:mrow><mml:msub><mml:mrow><mml:mover accent="true"><mml:mi>r</mml:mi>
<mml:mo stretchy="true">^</mml:mo>
</mml:mover>
</mml:mrow>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>t</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
</mml:mrow>
</mml:math>
</inline-formula>
is the
instantaneous firing rate of the recording site <italic>i</italic>
, <italic>C</italic>
<italic><sub>i</sub>
</italic>
is a
normalization coefficient. <italic>L</italic>
and <italic>R</italic>
indicate
signals pertaining respectively to the left and the right wheel. In absence of
neuronal activity, the robot goes straight with a constant angular speed of 5
rad/s that corresponds to a linear velocity of 16 cm/s. The coefficients <italic>C</italic>
<sub><italic>i</italic>
</sub>
can be computed
according to different criteria: they represent an estimate of the strength of
the connection between the corresponding input and output site (computed by
means of a linear regression), as reported in (Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]), or they simply represent the inverse of the
estimation of the maximum value that can be reached by the instantaneous firing
rate on each group (left versus right) of electrodes. The first method is
usually applied when decoding the activity of each unit in large population of
neurons (Georgopoulos et al. [<xref ref-type="bibr" rid="B14">29</xref>
]). For the experiments
reported here, we adopted the second: the used algorithm already selects
input-output pathways characterized by the maximal strength of the functional
connection and we only need to equalize them, in other words, when the robot is
far from obstacles the spontaneous activity should not cause the robot turning
preferentially clockwise or counterclockwise. Assuming that neurons on the left
and right sides are mostly excitatory, the minus sign in the control law allows
us to implement inhibitory contralateral connections between inputs and
outputs.</p>
<p>The WTA strategy has proven to be a more appropriate
decoding scheme than those already presented in our previous work (Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]). In fact, even though the WTA mechanism may result
in movements that are less smooth, the lowest values chosen for the angular
speed (i.e., 2 or 5 rad/s instead of 10 rad/s) facilitate the robot rotation
without affecting the general behavior and, as a consequence, the robot can
reverse direction in much less space, actually realizing what the “brain” is
ordering to its “body.” This strategy is also suggested by the nonlinearity
of the IR sensors (for further details see the “Khepera II-IR sensors
report,” <ext-link ext-link-type="uri" xlink:href="http://ftp.k-team.com/khepera/documentation/Kh2IRAN.pdf">http://ftp.k-team.com/khepera/documentation/Kh2IRAN.pdf</ext-link>
) of the robot
that are capable to reliably detect an obstacle only when the robot is closer
than about 5 cm from an obstacle.</p>
</statement>
</sec>
<sec sec-type="subsection" id="subsec2.3"><title>2.3. Neural preparation and control architecture</title>
<statement id="head4"><title>Neural preparation and electrophysiological set-up</title>
<p>Dissociated neurons in culture randomly rearrange in a
bidimensional structure and, once they have established synaptic connections, they
show spontaneous neural activity (starting from about 7 days in vitro DIVs)
that can be modulated by means of electrical stimulation (Maeda et al. [<xref ref-type="bibr" rid="B16">30</xref>
]; Jimbo et al. [<xref ref-type="bibr" rid="B17">31</xref>
]; Marom and
Shahaf [<xref ref-type="bibr" rid="B9">32</xref>
]). We used
dissociated cultures of cortical neurons, extracted from rat embryos (E18).
Using standard methods previously described (Martinoia et al. [<xref ref-type="bibr" rid="B49">33</xref>
]; Chiappalone et al. [<xref ref-type="bibr" rid="B15">34</xref>
]), cells
were plated on microelectrode arrays (MEAs) (Figures <xref ref-type="fig" rid="fig1">1</xref>
(a), <xref ref-type="fig" rid="fig1">1</xref>
(b)) with 60
TiN/SiN electrodes (diameter 30 <italic>μ</italic>
m,
interdistance 200 <italic>μ</italic>
m) arranged on
an 8 × 8 square grid.
Experiments were performed in the range 18–42 DIVs, when the neuronal network
reaches its “mature” state, consisting of synchronized clustered activity
with minute-to-minute fluctuations in the probability of firing (Marom and
Shahaf [<xref ref-type="bibr" rid="B9">32</xref>
]).</p>
<p>The experimental set-up is based on the MEA60 system
supplied by Multi Channel Systems (MCS, Reutlingen, Germany). The system is
constituted by the following elements: a neuronal preparation cultured over an
MEA, a mounting support with a 60-channel amplifier (gain 1200x), a home made
8-channel stimulus generator, to deliver both current and voltage desired
stimulating signals, an inverted optical microscope, connected to a CCD camera
(Hamamatsu, Japan), to visually monitor the cells during the experiment, an
antivibration table and a Faraday cage.</p>
<p>Raw data are also monitored and recorded by using the
commercial software MCRack (Multi ChannelSystems, Reutlingen, Germany)
(sampling frequency was set to 10 kHz/channel). To confirm real-time behaviour,
neural data were also processed offline by using ad-hoc developed software tools
(Vato et al. [<xref ref-type="bibr" rid="B7">35</xref>
]; Chiappalone et al. [<xref ref-type="bibr" rid="B15">34</xref>
]).</p>
</statement>
<statement id="head5"><title>Control architecture</title>
<p>The control architecture presented in our previous
work (Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]) has partially changed and
some feature has been added. In particular, we have moved from xPC-Target
(<ext-link ext-link-type="uri" xlink:href="http://www.mathworks.com/products/xpctarget">http://www.mathworks.com/products/xpctarget</ext-link>
) that was not able to handle and
log the very large amount of data coming from neuronal network, to the QNX 6.1
(QNX software systems), a POSIX-compatible operating system specific for hard
real-time applications.</p>
<p>The present architecture involves three PCs. PC1 (P4,
2.8 GHz, 512 MB RAM), that is, the one that runs QNX, is equipped with an A/D
board PCI-6071E (National Instruments, Texas, USA) and it is responsible for
(a) electrophysiological signals acquisition, (b) online spike detection and
artifact blanking, (c) decoding of the spike trains, (d) handling the serial
communication with the robot and PC3, (e) coding of robot's proximity sensors
signals, (f) production of the pattern of stimuli that trigger the electrical
stimulator, and (g) experimental data logging (spikes, sensors activity, wheel
speeds, stimuli, robot trajectory). These tasks are processed by different
threads at different sampling rates, in particular tasks (a)-(b) are at 10 kHz,
tasks (c)–(e)-(f)-(g) are at 250 Hz, and task (d) is at 10 Hz.</p>
<p>A second computer, PC2 (P4, 2.8 GHz, 512 MB RAM,
Win2000), connected to PC1 through an Ethernet link, is the experimental
front-end. We used Simulink/Real-Time Workshop (the MathWorks) and the RT-Lab
package (Opal-RT) as development environments. This package generates two
processes that are executed in real-time on the target node, that is, PC1. This
system allows simultaneous acquisition of neural signals from up to 32
recording sites.</p>
<p>PC3 (P4 2.8 GHz 512 MB RAM, QNX 6.1) is in charge of
real-time tracking of the robot movement and it is connected through a serial
cable to the main computational node of the architecture (PC1). A CCD camera
(DSE TCC5 with 1/3 CCD SONY SuperHAD) is placed 1.5 m above the central
position of the arena. This physical arrangement of the camera allows a good
resolution while minimizing distortion at the boundaries of the arena. A
frame-grabber (Arvoo Picasso PCI-2SQ) acquires samples at 5 Hz from the camera
with a resolution of 640 × 480 pixels. One
pixel on the CCD sensor corresponds to ∼ 2 mm, so that the arena is contained in a 400 × 400 subwindow at
the center of the image. The detection phase is performed in a small square
portion (50 × 50 pixels) of the global field of the CCD camera (the
detection in such small area is low demanding in computing performance and thus
the process can be performed in real-time). This square area represents the
predicted robot location. A red round marker placed onto the top of the robot
and the detection of its position is based on a local evaluation of the
intensity of the RGB values of every pixel belonging to the detection window.</p>
</statement>
</sec>
<sec sec-type="subsection" id="subsec2.4"><title>2.4. Data analysis</title>
<p>Processing of electrophysiological signals</p>
<p> <italic>Poststimulus Time Histogram</italic>
. To investigate the neural activity evoked by
stimulation, we computed the poststimulus time histogram (i.e., PSTH), which
represents the impulse response of each site of the neural preparation to
electrical stimulation. The PSTHs were calculated by taking 400 ms time windows
from the recordings that follow each stimulus. We then counted the number of
spikes occurring in a 2–4 msec bin and divided this measure by the number of
stimuli (Rieke et al. [<xref ref-type="bibr" rid="B36">36</xref>
]). For our cultures, typical
PSTHs show an “early” (less than 50 msec) and a “late” (50–250
milliseconds) component (Shahaf and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]; Marom and Shahaf [<xref ref-type="bibr" rid="B9">32</xref>
]; Cozzi et al. [<xref ref-type="bibr" rid="B3">22</xref>
]).</p>
<p> <italic>Stimulus-Triggered Speed</italic>
. The stimulus-triggered speed (STS) is constructed by
averaging the speed waveform due to each stimulus. The robot has two
independent wheels, whose speeds are proportional to the neuronal activity of
two “brain” regions (i.e., the two set of electrodes selected as motor areas
within the network). These regions receive sensory feedback by two independent
stimulating sites. It is possible to construct two STSs for each input-output,
that is, sensory-motor pathways, for a total of 4 curves (i.e., the variations
of the speed of the left and the right wheels in response to the left stimulus
are the first two curves and constitute the first STS, and the variation of the
speed of the left and the right wheels in response to the right stimulus
constitute the second STS with the latter two curves), and it is also possible
to study the performance of the robot behavior by studying the side-selectivity
of the relationship between stimuli and speeds.</p>
<statement id="head6"><title>Indicators of robot performance</title>
<p>The behavior of the neurorobotic system can be
evaluated by means of the <italic>stimulus-triggered speed (STS)</italic>
, that is, the
average motor commands elicited by a single electrical
stimulus:<disp-formula id="eq2"><label>(2)</label>
<mml:math id="M8"><mml:mrow><mml:mtext>STS</mml:mtext>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>τ</mml:mi>
<mml:mo stretchy="false">)</mml:mo>
<mml:mo>=</mml:mo>
<mml:msub><mml:mrow><mml:mo stretchy="false">〈</mml:mo>
<mml:mi>ω</mml:mi>
<mml:mo stretchy="false">(</mml:mo>
<mml:mi>τ</mml:mi>
<mml:mo>−</mml:mo>
<mml:msub><mml:mi>t</mml:mi>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo stretchy="false">)</mml:mo>
<mml:mo>〉</mml:mo>
</mml:mrow>
<mml:mi>i</mml:mi>
</mml:msub>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:math>
</disp-formula>
where <italic>t</italic>
<italic><sub>i</sub>
</italic>
is the time
instant of delivery of the <italic>i</italic>
th stimulus and <italic>τ</italic>
is the time
coordinate.</p>
<p>In order to have a more general evaluation of the
robot performance during each trail (5′), we also used the following three parameters:
<list list-type="roman-lower"><list-item><p><italic>number of
hits</italic>
,</p>
</list-item>
<list-item><p><italic>trajectory
length</italic>
,</p>
</list-item>
<list-item><p><italic>space
covered</italic>
. The percentage of the arena area covered by the robot
path:
<disp-formula id="eq3"><label>(3)</label>
<mml:math id="M9"><mml:mrow><mml:mtext>SC</mml:mtext>
<mml:mo>=</mml:mo>
<mml:mfrac><mml:mrow><mml:mtext>NP</mml:mtext>
<mml:mo>⋅</mml:mo>
<mml:mi>c</mml:mi>
<mml:msup><mml:mi>f</mml:mi>
<mml:mn>2</mml:mn>
</mml:msup>
</mml:mrow>
<mml:mrow><mml:msub><mml:mi>A</mml:mi>
<mml:mrow><mml:mtext>arena</mml:mtext>
</mml:mrow>
</mml:msub>
<mml:mo>−</mml:mo>
<mml:mi>n</mml:mi>
<mml:msub><mml:mi>A</mml:mi>
<mml:mrow><mml:mtext>obstacle</mml:mtext>
</mml:mrow>
</mml:msub>
</mml:mrow>
</mml:mfrac>
<mml:mo>⋅</mml:mo>
<mml:mn>100</mml:mn>
<mml:mi>%</mml:mi>
<mml:mo>,</mml:mo>
</mml:mrow>
</mml:math>
</disp-formula>
where NP is the number of pixels
covered by the robot, <italic>c</italic>
<italic>f</italic>
<sup><italic>2</italic>
</sup>
is the area of
one pixel, <italic>n</italic>
is the number
of obstacles in the playground, <italic>A</italic>
<sub><italic>arena</italic>
</sub>
and <italic>A</italic>
<sub><italic>obstacle</italic>
</sub>
are,
respectively, the areas of the playground and of each obstacle.</p>
</list-item>
</list>
The software tools for of-line signal processing aimed at the analysis of the behavior of the neurorobotic system were developed using Matlab 6.5 (the MathWorks).</p>
</statement>
</sec>
</sec>
<sec sec-type="section" id="sec3"><title>3. RESULTS</title>
<sec sec-type="subsection" id="subsec3.1"><title>3.1. Computing performance</title>
<p>The feedback loop computation time reached by our
final neurorobotic architecture is under 1 millisecond; therefore, the
real-time performance in the closed-loop system is compatible with the response
time (4 ms) of our neuronal model. This value includes the time needed for (I)
the electrophysiological signals acquisition, (II) the spike detection and the
artifact suppression, (III) decoding of neural activity, (IV) computing of the
speeds of the robot's wheels, and (V) coding of sensory feedback. The relative
computational loads for each block are displayed in <xref ref-type="fig" rid="fig3">Figure 3</xref>
(a): the most
time-consuming parts are those running at 10 kHz, for technical reasons, and
the blocks including sampling rate transitions, such as the interface with the
robot, with the CCD camera system and with the stimulator. In these
experiments, we used a robot with a standard RS232 interface that supports a
baud rate of 9600 bit/s. We expect that by including a more recent protocol
(e.g., USB2 or Firewire), the block would be less time consuming and would
assure better performance. Spike detection and artifact blanking are also
time-consuming due to the high dimension of the signals being processed. The
performances were evaluated by means of Simulink Profiler, reported
schematically in Figure 3(b).</p>
</sec>
<sec sec-type="subsection" id="subsec3.2"><title>3.2. Identification of input-output sites</title>
<p>In order to obtain a reactive behavior, we need the
network to respond soon after the feedback stimulation, that is, we need
input-output pathways characterized by a relatively early (up to 50 ms) and
sustained response meaning a “high strength” in the functional connectivity
(Shahaf and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]). If the network reacts to the sensory feedback and
the evoked electrophysiological response is characterized by a relatively long
activation phase (up to 200–300 ms), the robot would not be able to react to
the presence of an obstacle in 100 ms (i.e., the delay among successive serial
communications between the system and the robot). This is one of the reasons
why we need to accurately select the input-output pathways, beside the fact
that only low-frequency stimulation can be delivered for not fatiguing the
culture (Shahaf and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]; Eytan et al.
[<xref ref-type="bibr" rid="B43">37</xref>
]). We need
the stimulus-evoked response to be fast, prolonged, reliable, and therefore
effective for the entire duration of the experiments (i.e., all day long).</p>
<p>As already said, the general aim is to have a robot
that follows a specific task on the basis of the spontaneous/stimulated
electrophysiological activity shown by the neuronal culture. To this end, it is
a fundamental pre-requisite to characterize the collective activity of the
network that will be connected to the robot (i.e., analysis of both spontaneous
and stimulus evoked neuronal activity). This characterization phase is
necessary since the unstructured nature of the culture does not allow us to a
priori define the sensory and motor areas that will be connected with the
sensory and motor areas of the robot, as it happens with portion of tissue with
a well-defined sensorimotor architecture (Reger et al. [<xref ref-type="bibr" rid="B44">7</xref>
];
Karniel et al. [<xref ref-type="bibr" rid="B46">11</xref>
]).</p>
<p>Thus, the goal of the characterization phase is to
select those channels of the MEA to be used as sensory inputs (i.e., connected
to the robot's sensors) and motor outputs (i.e., connected to the robot's
wheels) of the biological network.</p>
<p>To test the response to stimulation from different
sites in different areas of the neuronal network, trains of 50 electrical
stimuli are delivered (1.5 V peak to peak-extracellular stimulation, 500 <italic>μ</italic>
s, and duty
cycle 50%). This procedure is repeated from at least 5 arbitrarily chosen
electrodes (Wagenaar et al. [<xref ref-type="bibr" rid="B10">38</xref>
]).</p>
<p>The poststimulus time histogram-(PSTH) (i.e., the average
number of spikes obtained in response to a stimulus, at different latencies) is
then used for quantifying the strength of connections between a specific
stimulating sites and all the other recording sites. It is the impulse response
(in terms of instantaneous firing rate) to a single test stimulus.</p>
<p>The algorithm for the selection of the output (motor)
and input (sensory) sites supplies the I/O pairs corresponding to maximum
selectivity and it is based on network effective functional interconnectivity.
The ideal case is described in the following: given two (or more) stimulating
channels (e.g., S1 and S2) and two groups of recording sites (e.g., R1 and R2),
the strength of the connectivity S1-R1 and S2-R2 is “high” and
simultaneously, the strength of the connectivity S1-R2, and S2-R1 is “low”
(i.e., good selectivity in input-output pairs). The described scheme
guarantees, somehow, that the responses in the two (groups of) recording sites
are different on the basis of the stimulating electrodes. Of course the above
is an ideal situation and, since the mean connectivity of the network is high,
also due to the high density of plated cells, it is hard to get highly specific
responses in the input-output pathways.</p>
<p>The methodology that we developed to make a selection
of the pathways is the “selectivity map” (see <xref ref-type="sec" rid="subsec3.3">Section 3.3</xref>
for a typical map).
Each dot represents the PSTH area at a specific recording site given that there
was a stimulation from a couple of stimulating sites. All the possible
input-output combinations are explored and only the pathways producing
responses lying more distant from the diagonal (i.e., closer to the axis) are
selected.</p>
<p>Those specific pathways (of sensory-motor activations)
can be then conveniently utilized for driving the robot and for implementing
simple reactive behaviors (e.g., obstacle avoidance), as presented in Section
3.1.</p>
</sec>
<sec sec-type="subsection" id="subsec3.3"><title>3.3. Example behaviors of the neurorobotic system</title>
<p>Once the role of the microelectrodes (i.e., selection
of input-output sites) has been decided, the experiment can start. A desired
result is achieved when an improvement of the robot behavior is confirmed by
possible modification of the neuronal network dynamics (i.e., adaptation).</p>
<p>Each experiment with the neurorobotic system is
usually divided into the following phases:
<list list-type="roman-lower"><list-item><p>spontaneous
activity recording (5 minutes);</p>
</list-item>
<list-item><p>preprocessing:
test stimulus from 8 channels (serial stimulation);</p>
</list-item>
<list-item><p>input-output
channel selection: at least 2 channels for input (sensors) and 2 channels for
output (motors).</p>
</list-item>
<list-item><p>closed–loop
experiment: Robot running (5 + 5 + 5 minutes):
<list list-type="alpha-lower"><list-item><p>free
running;</p>
</list-item>
<list-item><p>obstacle avoidance with the application of a learning protocol (when the robot
hits an obstacle, a conditioning stimulus at 20 Hz frequency is delivered from
the collision side). The learning protocol is based on what reported in
literature: only a few examples of learning (i.e., potentiation and depression)
have been demonstrated for dissociated neurons cultured over MEAs and all of
them are based on the application of trains of stimuli at “high” frequency
(Jimbo et al. [<xref ref-type="bibr" rid="B5">39</xref>
]; Jimbo et al. [<xref ref-type="bibr" rid="B2">40</xref>
]; Tateno and Jimbo [<xref ref-type="bibr" rid="B45">41</xref>
]; Bonifazi et al.
[<xref ref-type="bibr" rid="B41">42</xref>
];
Ruaro et al. [<xref ref-type="bibr" rid="B35">43</xref>
]). We have evidence that
similar protocols have the effects to mainly potentiate the network
electrophysiological response in terms of number of evoked spikes (Chiappalone
et al. [<xref ref-type="bibr" rid="B48">44</xref>
]);</p>
</list-item>
<list-item><p>free
running.</p>
</list-item>
</list>
</p>
</list-item>
<list-item><p>post-processing:
<list list-type="alpha-lower"><list-item><p>spontaneous activity recording (5′);</p>
</list-item>
<list-item><p>test stimulus from the two chosen
stimulating channels.</p>
</list-item>
</list>
</p>
</list-item>
</list>
To avoid manual removal of the robot and possible damage due to wheels' motors heating in case of a collision against an obstacle, a step-back mechanism (2 seconds backward movement with an angular speed of 5 rad/s) was implemented.</p>
<p>A user-friendly GUI allows (i) to select input and
output channels (i.e., recording and stimulation sites), (ii) to choose among
different coding and decoding strategies, and (iii) to change all the
experimental parameters (e.g., spike-detection threshold, maximum robot speed,
cut-off frequency of the filter for the estimation of neural activity, maximum
stimulation rate).</p>
<p>The signals obtained at different levels in the
bi-directional interface are represented in <xref ref-type="fig" rid="fig4">Figure 4</xref>
. The spike trains are then
low-pass filtered to obtain instantaneous firing rates that are considered as
indicators of the level of neural activity. The cutoff frequency of the filter
is set, in this case, at 1 Hz. The values corresponding to reasonable response
times range from 0.5 to 1 Hz: in fact they represent a good compromise between
fast response and time integration requirements. The previously adopted cutoff
frequency of 0.1 Hz (Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]) was not suitable for bursting networks because the
long-term effect of time integration lets the past events (previous bursts) to
weight more than instantaneous activity.</p>
<p>The motor commands (i.e., the speeds) are then
obtained according to the control law which implements inhibitory
contro-lateral connections between inputs and outputs (see decoding schemes
section for details), thus we expect that a feedback stimulus coming from left
sensors would result in a decrease of the speed of the right wheel of the
robot, and a stimulus coming from right sensors would determine a deceleration
of the speed of the robot's left wheel. The proximity signals coming from the
sensors placed on the two sides of the robot are averaged in order to obtain
two feedback signals, each of them related to one side of the robot. <xref ref-type="fig" rid="fig4">Figure 4</xref>
shows the patterns of stimuli obtained from the feedback signals, according to
the binary coding scheme.</p>
<p>A number of preliminary experiments were performed
using, respectively, 2 stimulation and 2 recording sites. In the following, the
results of two example experiments are reported to let the reader better
understand the experimental procedure and appreciate the performances of the
developed closed-loop system.</p>
<p><xref ref-type="fig" rid="fig5">Figure 5</xref>
shows the PSTHs obtained during the characterization
phase of one example experiment. The responses evoked from different
stimulation sites are similar (i.e., Figures <xref ref-type="fig" rid="fig5">5(a)</xref>
and <xref ref-type="fig" rid="fig5">5(b)</xref>
), thus revealing a
low degree of selectivity and a high degree of connectivity. In a case like the
one presented in <xref ref-type="fig" rid="fig5">Figure 5</xref>
, the preparation can hardly be used to control the
robot and it is discarded.</p>
<p><xref ref-type="fig" rid="fig6">Figure 6</xref>
presents an example of good connectivity maps
obtained during the characterization phase (<xref ref-type="fig" rid="fig6">Figure 6</xref>
(a)) and after the robotic
experiment (<xref ref-type="fig" rid="fig6">Figure 6</xref>
(b)): the electrodes 15 and 45, that will be further chosen
as recording electrodes, are positioned close to the axis, indicating that
their responses to the stimulating channel are selective (see <xref ref-type="sec" rid="subsec3.2">Section 3.2</xref>
for
further details).</p>
<p>Figures <xref ref-type="fig" rid="fig7">Figure7(a)</xref>
and <xref ref-type="fig" rid="fig7">Figure7(b)</xref>
show the PSTHs corresponding to
the inputs/outputs chosen after the characterization phase; during an
experimental session with the robot (i.e., experiment is different from the
previous one). In this example, the recording electrode 15 is very sensitive to
the stimulation delivered from electrode 16 (top left) while it is quite
unaffected by stimuli delivered form electrode 48 (top right). At the same
time, the recording site 45 is not sensitive to stimuli coming from electrode
16 (bottom left) while it is very affected by stimulation from electrode 48. In
this case, different stimulation sites evoke very different response, thus
revealing a high degree of selectivity that is also confirmed by the
connectivity maps presented in <xref ref-type="fig" rid="fig6">Figure 6</xref>
.</p>
<p>The shapes of the PSTHs must be similar to those of
the PSTHs obtained during the characterization phase in order to ensure the
stability of the response of the neuronal culture. If the area of the PSTHs
drastically decreases at the end of the closed loop phases it means that the
neuronal network has been fatigued by excessive repeated stimulations (Shahaf
and Marom [<xref ref-type="bibr" rid="B25">5</xref>
]). The
wellness and stability of the culture are “sine qua non” conditions to be
verified before describing the neurorobotic behavior by means of the robot's
performance indicators. Under these conditions, the performance of the neural
preparation in controlling the robot can be represented by the STSs, depicted
in Figures <xref ref-type="fig" rid="fig7">Figure 7(c)</xref>
and <xref ref-type="fig" rid="fig7">Figure 7(d)</xref>
.</p>
<p>Examples of the robot trajectories are presented in
Figures <xref ref-type="fig" rid="fig8">Figure 8(a)</xref>
and <xref ref-type="fig" rid="fig8">Figure 8(b)</xref>
. <xref ref-type="fig" rid="fig8">Figure 8(c)</xref>
shows the indicators generally used for
quantifying the robot performance. The first indicator alone, that is, the
number of hits, is not sufficient for describing the performances of an
obstacle avoidance task. In fact, a low number of hits could result from
limited robot movements or from the repetition of the same trajectory. For this
reason, it is necessary to consider also the fraction of space covered by the
robot and the length of its trajectory. Together, these simple indicators
evaluate the robot performances inside the arena, even if they are not related
to the sensory feedback coming from the external environment. These parameters
do not allow quantifying any relationship between the motor response and the
sensory information, but, considering different phases, if the robot covered
the same area and the trajectories are in the same order, then the two phases
are comparable, and a reduction of the number of hits should indicate an
improvement of the robot's behavior. An improvement in the robot's behavior
must correspond to an improvement in the relationship between the motor
response and the feedback sensory information (i.e., the STSs). The STS is the
only parameter that permits to understand and demonstrate whether a different
behavior of the robot actually corresponds to a different dynamics of the
neuronal activity, and for this reason it can be considered the best indicator
of the performance of the overall neurorobotic system.</p>
<p>The comparison of the STSs and connectivity maps
obtained during each phase illustrates that a modification in the robot's
behavior has occurred. Therefore, one could speculate that the origin of such a
modification relies on specific synaptic changes, (i.e., Hebbian potentiation
in terms of number of evokes spikes) of the neurons placed at the recording
electrodes (Jimbo et al.[<xref ref-type="bibr" rid="B5">39</xref>
]; (Jimbo et al. [<xref ref-type="bibr" rid="B2">40</xref>
]; Tateno and Jimbo [<xref ref-type="bibr" rid="B45">41</xref>
]; Bonifazi et al.
[<xref ref-type="bibr" rid="B41">42</xref>
];
Ruaro et al. [<xref ref-type="bibr" rid="B35">43</xref>
]). We cannot infer or
demonstrate that synaptic changes are pathways specific but considering the
global behavior of the recording sites after a neurorobotic experiments a
possible effect at (sub)population level (i.e., a kind of network plasticity)
has occurred induced by the external correlated stimulation (Chiappalone
et al. [<xref ref-type="bibr" rid="B48">44</xref>
]).</p>
</sec>
</sec>
<sec sec-type="section" id="sec4"><title>4. DISCUSSION AND CONCLUSION</title>
<p>“We have this common internal neural language that we
are born with and so if you can exploit that with the right stimuli then you
are going to help the brain develop to do the things like reason.” (Shaw
[<xref ref-type="bibr" rid="B38">45</xref>
].)</p>
<p>We have developed a general real-time, bidirectional
neural interface platform. The system is capable of acquiring multisite
electrophysiological activity at 10 kHz per channel, to perform spike detection
and artifact suppression, from up to 32 channels. That is a step forward with respect
to simpler systems (Reger et al.[<xref ref-type="bibr" rid="B44">7</xref>
]; Karniel et al.[<xref ref-type="bibr" rid="B46">11</xref>
]) or to bidirectional systems implemented by others
(DeMarse et al.[<xref ref-type="bibr" rid="B47">12</xref>
]; Bakkum et al. [<xref ref-type="bibr" rid="B12">13</xref>
]) or by ourselves (Martinoia et al. [<xref ref-type="bibr" rid="B37">14</xref>
]; Cozzi et al.[<xref ref-type="bibr" rid="B39">18</xref>
]).</p>
<p>The use of portions of brain, such as the brainstem of
a sea lamprey, connected to an artificial device represents the very first
application of the “embodied electrophysiology” paradigm. The main difference
between the lamprey-based neurorobotic system and the bidirectional interface
we developed is the fact that in the lamprey preparation the circuitry
governing the stabilization and orientation during swimming (Deliagina
[<xref ref-type="bibr" rid="B1">46</xref>
]; Deliagina et al. [<xref ref-type="bibr" rid="B50">47</xref>
]) maintains the original citoarchitecture and the natural
input-output system is used as a controller to drive the robotic body
(Reger
et al. [<xref ref-type="bibr" rid="B44">7</xref>
]; Karniel et al.
[<xref ref-type="bibr" rid="B46">11</xref>
]).</p>
<p>Potter and colleagues (DeMarse et al. [<xref ref-type="bibr" rid="B47">12</xref>
]; Bakkum et
al.
[<xref ref-type="bibr" rid="B12">13</xref>
]) overcame
the simplification of an already structured portion of brain presenting an
innovative embodied electrophysiology paradigm in which a randomly grown
neuronal networks controlled a simulated body. As they reported, it was simply
a “neuroethology experiment” to merely observe the effect of feedback
stimulation on the general behavior and where their “animat” has not to
perform any particular task. Our neurorobotic interface, on the contrary, has
to work in a reactive manner expressing a kind of “stimulus-driven behavior.”
Here we reported, in details, the methodology of the system (including hardware
and software features), an optimized way to identify the “functional I/O
pathways” and we presented a method to analyze the robot behavior and
correlate it to the network electrophysiology.</p>
<p>As described in the previous sections, in our system,
the spike trains can be decoded into motor commands (at 250 Hz) through a
variety of decoding strategies. Such motor commands are then used to control a
mobile robot, to which a specific task is assigned. Conversely, sensory signals
can be coded into patterns of stimulation (again, according to a variety of
coding schemes) and sent to a custom electrical stimulator with up to 8
stimulation channels. Although the algorithms used for spike detection and
artifact blanking are simple, compared to those adopted in other experimental
frameworks (Wagenaar and Potter [<xref ref-type="bibr" rid="B4">48</xref>
]; Obeid and Wolf [<xref ref-type="bibr" rid="B26">49</xref>
]), they allow a good level of reliability with the
advantage of an extremely light computational load. It should be underlined
that the presented experimental paradigm can be extended to other computational
schemes and one of the key features of the system is to allow testing different
coding and decoding strategies in relationship with optimal coding and
performances and with the capability of the neuronal system to adapt for a new
task in an actual closed loop environment.</p>
<p>The software architecture is flexible and modular, and
allows fast prototyping of new modules according to the experimental
requirements. Real-time performance was very good and it is comparable to other
simpler systems (in terms of recording/stimulating channels), previously
described in the literature (Reger et al.
[<xref ref-type="bibr" rid="B44">7</xref>
];
Martinoia et al. [<xref ref-type="bibr" rid="B37">14</xref>
]; Wagenaar and Potter [<xref ref-type="bibr" rid="B10">38</xref>
]; Cozzi et al. [<xref ref-type="bibr" rid="B39">18</xref>
]; Karniel et al.
[<xref ref-type="bibr" rid="B46">11</xref>
];
Wagenaar et al. [<xref ref-type="bibr" rid="B8">50</xref>
]). We developed a library of
coding and decoding modules, which, as mentioned, could be easily extended.
This is a key point with respect to possible implication for the development of
novel brain-machine interface with enhanced capabilities and bidirectional
interactivity.</p>
<p>Concerning our particular application, we also
developed the tools for studying the ability of a culture of cortical neurons
to process information in order to drive a robot according to a defined motor
task (with a particular emphasis on the method for input-output pathways
selection), and at the same time it allows to supervise the population activity
changes in response to external feedbacks. It should be stressed that this is
the first time that a closed-loop neurorobotic system (with in vitro neuronal
populations) is utilized for performing specific behavioral oriented tasks.</p>
<p>The proposed experimental framework creates new
possibility for investigating basic mechanisms of learning and adaptation
(e.g., distributed synaptic plasticity, long term potentiation (LTP) and long
term depression (LTD)) by directly studying how behavior arises from the
emerging collective dynamics of a neuronal ensemble. Additionally, the
experimental system could be also conveniently utilized and adapted to other in
vitro models such as acute, organotypic slices, and patterned neurons, where
the network architecture is partly preserved or can be designed.</p>
<p>Finally, on a long term perspective, this approach
could have a relevant impact in the field of bio-inspired computational systems
and for the development of novel brain-computer interfaces and of advanced
neuroprosthetic devices.</p>
</sec>
</body>
<back><ack><title>ACKNOWLEDGMENTS</title>
<p>The authors wish to thank Dr. M. Tedesco (Brunella)
for cell culture preparation. This work is partly supported by the EU Grant
IST-2001-33564 (Neurobit), and by a grant from the Italian Ministry of
University, Education and Research. Novellino's Ph.D. fellowship was supported
by ett s.r.l. (<ext-link ext-link-type="uri" xlink:href="http://www.ettsolutions.com">http://www.ettsolutions.com</ext-link>
).</p>
</ack>
<ref-list><ref id="B46"><label>1</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wagenaar</surname>
<given-names>DA</given-names>
</name>
<name><surname>Pine</surname>
<given-names>J</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>Effective parameters for stimulation of dissociated cultures using multi-electrode arrays</article-title>
<source><italic>Journal of Neuroscience Methods</italic>
</source>
<year>2004</year>
<volume>138</volume>
<issue>1-2</issue>
<fpage>27</fpage>
<lpage>37</lpage>
<pub-id pub-id-type="pmid">15325108</pub-id>
</citation>
</ref>
<ref id="B40"><label>2</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Sharp</surname>
<given-names>AA</given-names>
</name>
<name><surname>O'Neil</surname>
<given-names>MB</given-names>
</name>
<name><surname>Abbott</surname>
<given-names>LF</given-names>
</name>
<name><surname>Marder</surname>
<given-names>E</given-names>
</name>
</person-group>
<article-title>The dynamic clamp: artificial conductances in biological neurons</article-title>
<source><italic>Trends in Neurosciences</italic>
</source>
<year>1993</year>
<volume>16</volume>
<issue>10</issue>
<fpage>389</fpage>
<lpage>394</lpage>
<pub-id pub-id-type="pmid">7504352</pub-id>
</citation>
</ref>
<ref id="B22"><label>3</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Le Masson</surname>
<given-names>G</given-names>
</name>
<name><surname>Renaud-Le Masson</surname>
<given-names>S</given-names>
</name>
<name><surname>Debay</surname>
<given-names>D</given-names>
</name>
<name><surname>Bal</surname>
<given-names>T</given-names>
</name>
</person-group>
<article-title>Feedback inhibition controls spike transfer in hybrid thalamic circuits</article-title>
<source><italic>Nature</italic>
</source>
<year>2002</year>
<volume>417</volume>
<issue>6891</issue>
<fpage>854</fpage>
<lpage>858</lpage>
<pub-id pub-id-type="pmid">12075353</pub-id>
</citation>
</ref>
<ref id="B48"><label>4</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wagenaar</surname>
<given-names>DA</given-names>
</name>
<name><surname>Madhavan</surname>
<given-names>R</given-names>
</name>
<name><surname>Pine</surname>
<given-names>J</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>Controlling bursting in cortical cultures with closed-loop multi-electrode stimulation</article-title>
<source><italic>Journal of Neuroscience</italic>
</source>
<year>2005</year>
<volume>25</volume>
<issue>3</issue>
<fpage>680</fpage>
<lpage>688</lpage>
<pub-id pub-id-type="pmid">15659605</pub-id>
</citation>
</ref>
<ref id="B39"><label>5</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Shahaf</surname>
<given-names>G</given-names>
</name>
<name><surname>Marom</surname>
<given-names>S</given-names>
</name>
</person-group>
<article-title>Learning in networks of cortical neurons</article-title>
<source><italic>Journal of Neuroscience</italic>
</source>
<year>2001</year>
<volume>21</volume>
<issue>22</issue>
<fpage>8782</fpage>
<lpage>8788</lpage>
<pub-id pub-id-type="pmid">11698590</pub-id>
</citation>
</ref>
<ref id="B27"><label>6</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Marom</surname>
<given-names>S</given-names>
</name>
<name><surname>Eytan</surname>
<given-names>D</given-names>
</name>
</person-group>
<article-title>Learning in ex-vivo developing networks of cortical neurons</article-title>
<source><italic>Progress in Brain Research</italic>
</source>
<year>2005</year>
<volume>147</volume>
<fpage>189</fpage>
<lpage>199</lpage>
<pub-id pub-id-type="pmid">15581706</pub-id>
</citation>
</ref>
<ref id="B35"><label>7</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Reger</surname>
<given-names>BD</given-names>
</name>
<name><surname>Fleming</surname>
<given-names>KM</given-names>
</name>
<name><surname>Sanguineti</surname>
<given-names>V</given-names>
</name>
<name><surname>Alford</surname>
<given-names>S</given-names>
</name>
<name><surname>Mussa-Ivaldi</surname>
<given-names>FA</given-names>
</name>
</person-group>
<article-title>Connecting brains to robots: an artificial body for studying the computational properties of neural tissues</article-title>
<source><italic>Artificial Life</italic>
</source>
<year>2000</year>
<volume>6</volume>
<issue>4</issue>
<fpage>307</fpage>
<lpage>324</lpage>
<pub-id pub-id-type="pmid">11348584</pub-id>
</citation>
</ref>
<ref id="B50"><label>8</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wessberg</surname>
<given-names>J</given-names>
</name>
<name><surname>Stambaugh</surname>
<given-names>CR</given-names>
</name>
<name><surname>Kralik</surname>
<given-names>JD</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Real-time prediction of hand trajectory by ensembles of cortical neurons in primates</article-title>
<source><italic>Nature</italic>
</source>
<year>2000</year>
<volume>408</volume>
<issue>6810</issue>
<fpage>361</fpage>
<lpage>365</lpage>
<pub-id pub-id-type="pmid">11099043</pub-id>
</citation>
</ref>
<ref id="B32"><label>9</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Nicolelis</surname>
<given-names>MAL</given-names>
</name>
</person-group>
<article-title>Actions from thoughts</article-title>
<source><italic>Nature</italic>
</source>
<year>2001</year>
<volume>409</volume>
<issue>6818</issue>
<fpage>403</fpage>
<lpage>407</lpage>
<pub-id pub-id-type="pmid">11201755</pub-id>
</citation>
</ref>
<ref id="B38"><label>10</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Schwartz</surname>
<given-names>AB</given-names>
</name>
<name><surname>Taylor</surname>
<given-names>DM</given-names>
</name>
<name><surname>Tillery</surname>
<given-names>SIH</given-names>
</name>
</person-group>
<article-title>Extraction algorithms for cortical control of arm prosthetics</article-title>
<source><italic>Current Opinion in Neurobiology</italic>
</source>
<year>2001</year>
<volume>11</volume>
<issue>6</issue>
<fpage>701</fpage>
<lpage>708</lpage>
<pub-id pub-id-type="pmid">11741021</pub-id>
</citation>
</ref>
<ref id="B21"><label>11</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Karniel</surname>
<given-names>A</given-names>
</name>
<name><surname>Kositsky</surname>
<given-names>M</given-names>
</name>
<name><surname>Fleming</surname>
<given-names>KM</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Computational analysis in vitro: dynamics and plasticity of a neuro-robotic system</article-title>
<source><italic>Journal of Neural Engineering</italic>
</source>
<year>2005</year>
<volume>2</volume>
<issue>3</issue>
<fpage>S250</fpage>
<lpage>S265</lpage>
<pub-id pub-id-type="pmid">16135888</pub-id>
</citation>
</ref>
<ref id="B13"><label>12</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>DeMarse</surname>
<given-names>TB</given-names>
</name>
<name><surname>Wagenaar</surname>
<given-names>DA</given-names>
</name>
<name><surname>Blau</surname>
<given-names>AW</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>The neurally controlled animat: biological brains acting with simulated bodies</article-title>
<source><italic>Autonomous Robots</italic>
</source>
<year>2001</year>
<volume>11</volume>
<issue>3</issue>
<fpage>305</fpage>
<lpage>310</lpage>
</citation>
</ref>
<ref id="B1"><label>13</label>
<citation citation-type="confproc"><person-group person-group-type="author"><name><surname>Bakkum</surname>
<given-names>DJ</given-names>
</name>
<name><surname>Shkolnik</surname>
<given-names>AC</given-names>
</name>
<name><surname>Ben-Ary</surname>
<given-names>G</given-names>
</name>
<name><surname>Gamblen</surname>
<given-names>P</given-names>
</name>
<name><surname>DeMarse</surname>
<given-names>TB</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>Removing some ‘A’ from AI: embodied cultured networks</article-title>
<conf-name>International Seminar on Embodied Artificial Intelligence, vol.3139</conf-name>
<conf-date>July 2004</conf-date>
<conf-loc>Dagstuhl Castle, Germany</conf-loc>
<fpage>130</fpage>
<lpage>145</lpage>
<series>Lecture Notes in Artificial Intelligence</series>
</citation>
</ref>
<ref id="B29"><label>14</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Martinoia</surname>
<given-names>S</given-names>
</name>
<name><surname>Sanguineti</surname>
<given-names>V</given-names>
</name>
<name><surname>Cozzi</surname>
<given-names>L</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Towards an embodied in vitro electrophysiology:the NeuroBIT project</article-title>
<source><italic>Neurocomputing</italic>
</source>
<year>2004</year>
<volume>58–60</volume>
<fpage>1065</fpage>
<lpage>1072</lpage>
</citation>
</ref>
<ref id="B34"><label>15</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>Chapter 4 distributed processing in cultured neuronal networks</article-title>
<source><italic>Progress in Brain Research</italic>
</source>
<year>2001</year>
<volume>130</volume>
<fpage>49</fpage>
<lpage>62</lpage>
<pub-id pub-id-type="pmid">11480288</pub-id>
</citation>
</ref>
<ref id="B30"><label>16</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Mussa-Ivaldi</surname>
<given-names>FA</given-names>
</name>
<name><surname>Miller</surname>
<given-names>LE</given-names>
</name>
</person-group>
<article-title>Brain-machine interfaces: computational demands and clinical needs meet basic neuroscience</article-title>
<source><italic>Trends in Neurosciences</italic>
</source>
<year>2003</year>
<volume>26</volume>
<issue>6</issue>
<fpage>329</fpage>
<lpage>334</lpage>
<pub-id pub-id-type="pmid">12798603</pub-id>
</citation>
</ref>
<ref id="B31"><label>17</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Nicolelis</surname>
<given-names>MAL</given-names>
</name>
</person-group>
<article-title>Brain-machine interfaces to restore motor function and probe neural circuits</article-title>
<source><italic>Nature Reviews Neuroscience</italic>
</source>
<year>2003</year>
<volume>4</volume>
<issue>5</issue>
<fpage>417</fpage>
<lpage>422</lpage>
</citation>
</ref>
<ref id="B10"><label>18</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Cozzi</surname>
<given-names>L</given-names>
</name>
<name><surname>D'Angelo</surname>
<given-names>P</given-names>
</name>
<name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Coding and decoding of information in a bi-directional neural interface</article-title>
<source><italic>Neurocomputing</italic>
</source>
<year>2005</year>
<volume>65-66</volume>
<fpage>783</fpage>
<lpage>792</lpage>
</citation>
</ref>
<ref id="B25"><label>19</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Maes</surname>
<given-names>P</given-names>
</name>
</person-group>
<article-title>Modeling adaptive autonomous agents</article-title>
<source><italic>Artificial Life</italic>
</source>
<year>1994</year>
<volume>1</volume>
<issue>1-2</issue>
<fpage>135</fpage>
<lpage>162</lpage>
</citation>
</ref>
<ref id="B4"><label>20</label>
<citation citation-type="book"><person-group person-group-type="author"><name><surname>Brooks</surname>
<given-names>RA</given-names>
</name>
</person-group>
<source><italic>Robot: The Future of Flesh and Machines</italic>
</source>
<year>2002</year>
<publisher-loc>London, UK</publisher-loc>
<publisher-name>Allen Lane, The Penguin Press</publisher-name>
</citation>
</ref>
<ref id="B3"><label>21</label>
<citation citation-type="book"><person-group person-group-type="author"><name><surname>Braitenberg</surname>
<given-names>V</given-names>
</name>
</person-group>
<source><italic>Vehicles: Experiments in Synthetic Psychology</italic>
</source>
<year>1984</year>
<publisher-loc>Cambridge, Mass, USA</publisher-loc>
<publisher-name>The MIT Press</publisher-name>
</citation>
</ref>
<ref id="B9"><label>22</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Cozzi</surname>
<given-names>L</given-names>
</name>
<name><surname>D'Angelo</surname>
<given-names>P</given-names>
</name>
<name><surname>Sanguineti</surname>
<given-names>V</given-names>
</name>
</person-group>
<article-title>Encoding of time-varying stimuli in populations of cultured neurons</article-title>
<source><italic>Biological Cybernetics</italic>
</source>
<year>2006</year>
<volume>94</volume>
<issue>5</issue>
<fpage>335</fpage>
<lpage>349</lpage>
<pub-id pub-id-type="pmid">16479398</pub-id>
</citation>
</ref>
<ref id="B17"><label>23</label>
<citation citation-type="confproc"><person-group person-group-type="author"><name><surname>Grattarola</surname>
<given-names>M</given-names>
</name>
<name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<name><surname>Davide</surname>
<given-names>F</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Burst analysis of chemically stimulated spinal cord neuronal networks cultured on microelectrode arrays</article-title>
<conf-name>In Proceedings of the 23rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society, vol.1</conf-name>
<conf-date>October 2001</conf-date>
<conf-loc>Istanbul, Turkey</conf-loc>
<fpage>729</fpage>
<lpage>732</lpage>
</citation>
</ref>
<ref id="B6"><label>24</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Chapin</surname>
<given-names>JK</given-names>
</name>
<name><surname>Moxon</surname>
<given-names>KA</given-names>
</name>
<name><surname>Markowitz</surname>
<given-names>RS</given-names>
</name>
<name><surname>Nicolelis</surname>
<given-names>MAL</given-names>
</name>
</person-group>
<article-title>Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex</article-title>
<source><italic>Nature Neuroscience</italic>
</source>
<year>1999</year>
<volume>2</volume>
<issue>7</issue>
<fpage>664</fpage>
<lpage>670</lpage>
</citation>
</ref>
<ref id="B5"><label>25</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Carmena</surname>
<given-names>JM</given-names>
</name>
<name><surname>Lebedev</surname>
<given-names>MA</given-names>
</name>
<name><surname>Crist</surname>
<given-names>RE</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Learning to control a brain-machine interface for reaching and grasping by primates</article-title>
<source><italic>PLoS Biology</italic>
</source>
<year>2003</year>
<volume>1</volume>
<issue>2</issue>
<fpage>193</fpage>
<lpage>208</lpage>
</citation>
</ref>
<ref id="B49"><label>26</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wessberg</surname>
<given-names>J</given-names>
</name>
<name><surname>Nicolelis</surname>
<given-names>MAL</given-names>
</name>
</person-group>
<article-title>Optimizing a linear algorithm for real-time robotic control using chronic cortical ensemble recordings in monkeys</article-title>
<source><italic>Journal of Cognitive Neuroscience</italic>
</source>
<year>2004</year>
<volume>16</volume>
<issue>6</issue>
<fpage>1022</fpage>
<lpage>1035</lpage>
<pub-id pub-id-type="pmid">15298789</pub-id>
</citation>
</ref>
<ref id="B15"><label>27</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Fetz</surname>
<given-names>EE</given-names>
</name>
</person-group>
<article-title>Real-time control of a robotic arm by neuronal ensembles</article-title>
<source><italic>Nature Neuroscience</italic>
</source>
<year>1999</year>
<volume>2</volume>
<issue>7</issue>
<fpage>583</fpage>
<lpage>584</lpage>
</citation>
</ref>
<ref id="B23"><label>28</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Lebedev</surname>
<given-names>MA</given-names>
</name>
<name><surname>Nicolelis</surname>
<given-names>MAL</given-names>
</name>
</person-group>
<article-title>Brain-machine interfaces: past, present and future</article-title>
<source><italic>Trends in Neurosciences</italic>
</source>
<year>2006</year>
<volume>29</volume>
<issue>9</issue>
<fpage>536</fpage>
<lpage>546</lpage>
<pub-id pub-id-type="pmid">16859758</pub-id>
</citation>
</ref>
<ref id="B16"><label>29</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Georgopoulos</surname>
<given-names>AP</given-names>
</name>
<name><surname>Schwartz</surname>
<given-names>AB</given-names>
</name>
<name><surname>Kettner</surname>
<given-names>RE</given-names>
</name>
</person-group>
<article-title>Neuronal population coding on movement direction</article-title>
<source><italic>Science</italic>
</source>
<year>1986</year>
<volume>233</volume>
<issue>4771</issue>
<fpage>1416</fpage>
<lpage>1419</lpage>
<pub-id pub-id-type="pmid">3749885</pub-id>
</citation>
</ref>
<ref id="B24"><label>30</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Maeda</surname>
<given-names>E</given-names>
</name>
<name><surname>Kuroda</surname>
<given-names>Y</given-names>
</name>
<name><surname>Robinson</surname>
<given-names>HPC</given-names>
</name>
<name><surname>Kawana</surname>
<given-names>A</given-names>
</name>
</person-group>
<article-title>Modification of parallel activity elicited by propagating bursts in developing networks of rat cortical neurones</article-title>
<source><italic>European Journal of Neuroscience</italic>
</source>
<year>1998</year>
<volume>10</volume>
<issue>2</issue>
<fpage>488</fpage>
<lpage>496</lpage>
<pub-id pub-id-type="pmid">9749711</pub-id>
</citation>
</ref>
<ref id="B20"><label>31</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Jimbo</surname>
<given-names>Y</given-names>
</name>
<name><surname>Kawana</surname>
<given-names>A</given-names>
</name>
<name><surname>Parodi</surname>
<given-names>P</given-names>
</name>
<name><surname>Torre</surname>
<given-names>V</given-names>
</name>
</person-group>
<article-title>The dynamics of a neuronal culture of dissociated cortical neurons of neonatal rats</article-title>
<source><italic>Biological Cybernetics</italic>
</source>
<year>2000</year>
<volume>83</volume>
<issue>1</issue>
<fpage>1</fpage>
<lpage>20</lpage>
<pub-id pub-id-type="pmid">10933234</pub-id>
</citation>
</ref>
<ref id="B26"><label>32</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Marom</surname>
<given-names>S</given-names>
</name>
<name><surname>Shahaf</surname>
<given-names>G</given-names>
</name>
</person-group>
<article-title>Development, learning and memory in large random networks of cortical neurons: lessons beyond anatomy</article-title>
<source><italic>Quarterly Reviews of Biophysics</italic>
</source>
<year>2002</year>
<volume>35</volume>
<issue>1</issue>
<fpage>63</fpage>
<lpage>87</lpage>
<pub-id pub-id-type="pmid">11997981</pub-id>
</citation>
</ref>
<ref id="B28"><label>33</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Martinoia</surname>
<given-names>S</given-names>
</name>
<name><surname>Bonzano</surname>
<given-names>L</given-names>
</name>
<name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<name><surname>Tedesco</surname>
<given-names>M</given-names>
</name>
<name><surname>Marcoli</surname>
<given-names>M</given-names>
</name>
<name><surname>Maura</surname>
<given-names>G</given-names>
</name>
</person-group>
<article-title>In vitro cortical neuronal networks as a new high-sensitive system for biosensing applications</article-title>
<source><italic>Biosensors and Bioelectronics</italic>
</source>
<year>2005</year>
<volume>20</volume>
<issue>10</issue>
<fpage>2071</fpage>
<lpage>2078</lpage>
<pub-id pub-id-type="pmid">15741077</pub-id>
</citation>
</ref>
<ref id="B8"><label>34</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<name><surname>Bove</surname>
<given-names>M</given-names>
</name>
<name><surname>Vato</surname>
<given-names>A</given-names>
</name>
<name><surname>Tedesco</surname>
<given-names>M</given-names>
</name>
<name><surname>Martinoia</surname>
<given-names>S</given-names>
</name>
</person-group>
<article-title>Dissociated cortical networks show spontaneously correlated activity patterns during in vitro development</article-title>
<source><italic>Brain Research</italic>
</source>
<year>2006</year>
<volume>1093</volume>
<issue>1</issue>
<fpage>41</fpage>
<lpage>53</lpage>
<pub-id pub-id-type="pmid">16712817</pub-id>
</citation>
</ref>
<ref id="B43"><label>35</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Vato</surname>
<given-names>A</given-names>
</name>
<name><surname>Bonzano</surname>
<given-names>L</given-names>
</name>
<name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<etal></etal>
</person-group>
<article-title>Spike manager: a new tool for spontaneous and evoked neuronal networks activity characterization</article-title>
<source><italic>Neurocomputing</italic>
</source>
<year>2004</year>
<volume>58–60</volume>
<fpage>1153</fpage>
<lpage>1161</lpage>
</citation>
</ref>
<ref id="B36"><label>36</label>
<citation citation-type="book"><person-group person-group-type="author"><name><surname>Rieke</surname>
<given-names>F</given-names>
</name>
<name><surname>Warland</surname>
<given-names>D</given-names>
</name>
<name><surname>de Ruyter van Steveninck</surname>
<given-names>R</given-names>
</name>
<name><surname>Bialek</surname>
<given-names>W</given-names>
</name>
</person-group>
<source><italic>Spikes: Exploring the Neural Code</italic>
</source>
<year>1997</year>
<publisher-loc>Cambridge, Mass, USA</publisher-loc>
<publisher-name>The MIT Press</publisher-name>
</citation>
</ref>
<ref id="B14"><label>37</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Eytan</surname>
<given-names>D</given-names>
</name>
<name><surname>Brenner</surname>
<given-names>N</given-names>
</name>
<name><surname>Marom</surname>
<given-names>S</given-names>
</name>
</person-group>
<article-title>Selective adaptation in networks of cortical neurons</article-title>
<source><italic>Journal of Neuroscience</italic>
</source>
<year>2003</year>
<volume>23</volume>
<issue>28</issue>
<fpage>9349</fpage>
<lpage>9356</lpage>
<pub-id pub-id-type="pmid">14561862</pub-id>
</citation>
</ref>
<ref id="B45"><label>38</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wagenaar</surname>
<given-names>DA</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>A versatile all-channel stimulator for electrode arrays, with real-time control</article-title>
<source><italic>Journal of Neural Engineering</italic>
</source>
<year>2004</year>
<volume>1</volume>
<issue>1</issue>
<fpage>39</fpage>
<lpage>45</lpage>
<pub-id pub-id-type="pmid">15876621</pub-id>
</citation>
</ref>
<ref id="B18"><label>39</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Jimbo</surname>
<given-names>Y</given-names>
</name>
<name><surname>Robinson</surname>
<given-names>HPC</given-names>
</name>
<name><surname>Kawana</surname>
<given-names>A</given-names>
</name>
</person-group>
<article-title>Strengthening of synchronized activity by tetanic stimulation in cortical cultures: application of planar electrode arrays</article-title>
<source><italic>IEEE Transactions on Biomedical Engineering</italic>
</source>
<year>1998</year>
<volume>45</volume>
<issue>11</issue>
<fpage>1297</fpage>
<lpage>1304</lpage>
<pub-id pub-id-type="pmid">9805828</pub-id>
</citation>
</ref>
<ref id="B19"><label>40</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Jimbo</surname>
<given-names>Y</given-names>
</name>
<name><surname>Tateno</surname>
<given-names>T</given-names>
</name>
<name><surname>Robinson</surname>
<given-names>HPC</given-names>
</name>
</person-group>
<article-title>Simultaneous induction of pathway-specific potentiation and depression in networks of cortical neurons</article-title>
<source><italic>Biophysical Journal</italic>
</source>
<year>1999</year>
<volume>76</volume>
<issue>2</issue>
<fpage>670</fpage>
<lpage>678</lpage>
<pub-id pub-id-type="pmid">9929472</pub-id>
</citation>
</ref>
<ref id="B42"><label>41</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Tateno</surname>
<given-names>T</given-names>
</name>
<name><surname>Jimbo</surname>
<given-names>Y</given-names>
</name>
</person-group>
<article-title>Activity-dependent enhancement in the reliability of correlated spike timings in cultured cortical neurons</article-title>
<source><italic>Biological Cybernetics</italic>
</source>
<year>1999</year>
<volume>80</volume>
<issue>1</issue>
<fpage>45</fpage>
<lpage>55</lpage>
<pub-id pub-id-type="pmid">9951397</pub-id>
</citation>
</ref>
<ref id="B2"><label>42</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Bonifazi</surname>
<given-names>P</given-names>
</name>
<name><surname>Ruaro</surname>
<given-names>ME</given-names>
</name>
<name><surname>Torre</surname>
<given-names>V</given-names>
</name>
</person-group>
<article-title>Statistical properties of information processing in neuronal networks</article-title>
<source><italic>European Journal of Neuroscience</italic>
</source>
<year>2005</year>
<volume>22</volume>
<issue>11</issue>
<fpage>2953</fpage>
<lpage>2964</lpage>
<pub-id pub-id-type="pmid">16324130</pub-id>
</citation>
</ref>
<ref id="B37"><label>43</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Ruaro</surname>
<given-names>ME</given-names>
</name>
<name><surname>Bonifazi</surname>
<given-names>P</given-names>
</name>
<name><surname>Torre</surname>
<given-names>V</given-names>
</name>
</person-group>
<article-title>Toward the neurocomputer: image processing and pattern recognition with neuronal cultures</article-title>
<source><italic>IEEE Transactions on Biomedical Engineering</italic>
</source>
<year>2005</year>
<volume>52</volume>
<issue>3</issue>
<fpage>371</fpage>
<lpage>383</lpage>
<pub-id pub-id-type="pmid">15759567</pub-id>
</citation>
</ref>
<ref id="B7"><label>44</label>
<citation citation-type="confproc"><person-group person-group-type="author"><name><surname>Chiappalone</surname>
<given-names>M</given-names>
</name>
<name><surname>Massobrio</surname>
<given-names>P</given-names>
</name>
<name><surname>Tedesco</surname>
<given-names>M</given-names>
</name>
<name><surname>Martinoia</surname>
<given-names>S</given-names>
</name>
</person-group>
<article-title>Stimulus-induced synaptic changes in networks of cortical neurons</article-title>
<year>2006</year>
<conf-name>Proceedings of the 5th International Meeting on Substrate-Integrated Micro Electrode Arrays (SIMEA '06)</conf-name>
<conf-loc>Reutlingen, Germany</conf-loc>
<publisher-name>BIOPRO Baden-Wurttemberg GmbH</publisher-name>
<fpage>32</fpage>
<lpage>33</lpage>
</citation>
</ref>
<ref id="B41"><label>45</label>
<citation citation-type="book"><person-group person-group-type="author"><name><surname>Shaw</surname>
<given-names>GL</given-names>
</name>
</person-group>
<source><italic>Keeping Mozart in Mind</italic>
</source>
<year>2003</year>
<edition>2nd</edition>
<publisher-loc>San Diego, Calif, USA</publisher-loc>
<publisher-name>Academic Press</publisher-name>
</citation>
</ref>
<ref id="B11"><label>46</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Deliagina</surname>
<given-names>TG</given-names>
</name>
</person-group>
<article-title>Vestibular compensation in lampreys: impairment and recovery of equilibrium control during locomotion</article-title>
<source><italic>The Journal of Experimental Biology</italic>
</source>
<year>1997</year>
<volume>200</volume>
<issue>10</issue>
<fpage>1459</fpage>
<lpage>1471</lpage>
<pub-id pub-id-type="pmid">9319361</pub-id>
</citation>
</ref>
<ref id="B12"><label>47</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Deliagina</surname>
<given-names>TG</given-names>
</name>
<name><surname>Zelenin</surname>
<given-names>PV</given-names>
</name>
<name><surname>Fagerstedt</surname>
<given-names>P</given-names>
</name>
<name><surname>Grillner</surname>
<given-names>S</given-names>
</name>
<name><surname>Orlovsky</surname>
<given-names>GN</given-names>
</name>
</person-group>
<article-title>Activity of reticulospinal neurons during locomotion in the freely behaving lamprey</article-title>
<source><italic>Journal of Neurophysiology</italic>
</source>
<year>2000</year>
<volume>83</volume>
<issue>2</issue>
<fpage>853</fpage>
<lpage>863</lpage>
<pub-id pub-id-type="pmid">10669499</pub-id>
</citation>
</ref>
<ref id="B44"><label>48</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Wagenaar</surname>
<given-names>DA</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>Real-time multi-channel stimulus artifact suppression by local curve fitting</article-title>
<source><italic>Journal of Neuroscience Methods</italic>
</source>
<year>2002</year>
<volume>120</volume>
<issue>2</issue>
<fpage>113</fpage>
<lpage>120</lpage>
<pub-id pub-id-type="pmid">12385761</pub-id>
</citation>
</ref>
<ref id="B33"><label>49</label>
<citation citation-type="journal"><person-group person-group-type="author"><name><surname>Obeid</surname>
<given-names>I</given-names>
</name>
<name><surname>Wolf</surname>
<given-names>PD</given-names>
</name>
</person-group>
<article-title>Evaluation of spike-detection algorithms for a brain-machine interface application</article-title>
<source><italic>IEEE Transactions on Biomedical Engineering</italic>
</source>
<year>2004</year>
<volume>51</volume>
<issue>6</issue>
<fpage>905</fpage>
<lpage>911</lpage>
<pub-id pub-id-type="pmid">15188857</pub-id>
</citation>
</ref>
<ref id="B47"><label>50</label>
<citation citation-type="confproc"><person-group person-group-type="author"><name><surname>Wagenaar</surname>
<given-names>D</given-names>
</name>
<name><surname>DeMarse</surname>
<given-names>TB</given-names>
</name>
<name><surname>Potter</surname>
<given-names>SM</given-names>
</name>
</person-group>
<article-title>MeaBench: a toolset for multi-electrode data acquisition and on-line analysis</article-title>
<conf-name>Proceedings of the 2nd International IEEE EMBS Conference on Neural Engineering</conf-name>
<conf-date>March 2005</conf-date>
<conf-loc>Arlington, Va, USA</conf-loc>
<fpage>518</fpage>
<lpage>521</lpage>
</citation>
</ref>
</ref-list>
</back>
<floats-wrap><fig id="fig1" position="float"><label>Figure 1</label>
<caption><p>The main actors of the
neurorobotic set-up. (a) A commercial MEA by Multichannel Systems (Reutlingen,
Germany) with 60 electrodes. (b) Cortical neurons grow and develop a 2D network
over the MEA, in proximity of a recording microelectrode. (c) The Kephera robot
wandering in its arena, filled with cylindrical wooden obstacles.</p>
</caption>
<graphic xlink:href="CIN2007-12725.001"></graphic>
</fig>
<fig id="fig2" position="float"><label>Figure 2</label>
<caption><p>Computational
architecture of the closed-loop system. The signals coming from the infrared
sensors (IR) of the robot are translated into patterns of stimuli that are
delivered to the neural preparation through a set of selected stimulating
electrodes. Then the activity recorded by two groups of electrodes is evaluated
in terms of firing rate (i.e., mean number of detected spikes/s) and used as
driving speed for each of the robot's wheel.</p>
</caption>
<graphic xlink:href="CIN2007-12725.002"></graphic>
</fig>
<fig id="fig3" position="float"><label>Figure 3</label>
<caption><p>Computing performances in
the Simulink implementation of the neurorobotic interface. Different colors
correspond to different sampling rates (red = 10 kHz, green = 250 Hz, blue 5 Hz, yellow = mixed values),
whereas the numbers indicate signal dimensions (i.e., in this case, we had 2
inputs and 16 outputs). The percentages in each block indicate the relative
simulation time for each of the modules of the neurorobotic interface (2 inputs
and 16 outputs). (b) Library of the modules that can be used in the
neurorobotic interface. Each subset encloses Simulink blocks implementing
different algorithms for that purpose.</p>
</caption>
<graphic xlink:href="CIN2007-12725.003"></graphic>
</fig>
<fig id="fig4" position="float"><label>Figure 4</label>
<caption><p>Signals obtained at different levels of the
neurorobotic interface during 20 seconds of a free running session (c.f.
Results, for a detailed description of the experimental protocol). In this
particular experiment we used 2 stimulation sites and 2 recording sites. When
the robot is approaching an obstacle, the value of sensors increases and when
it overcomes a threshold (i.e., 500 levels) a feedback stimulus is delivered
(max frequency 1 Hz). The left and right detected spike trains are then
processed into motor commands, that is, left and right wheel speeds (the line
corresponds to 0 rad/s). The network was spontaneously active and during this phase
we recorded 2046 spikes from left and 1057 from the right one, resulting in a
mean wheel speed of 1.4 rad/s (left) and 1.6 rad/s (right).</p>
</caption>
<graphic xlink:href="CIN2007-12725.004"></graphic>
</fig>
<fig id="fig5" position="float"><label>Figure 5</label>
<caption><p>PSTHs of the evoked responses in a cortical neuronal
network. (a) The post stimulus time histograms obtained in all the responding
channels are reported over an 8 × 8 grid (i.e.,
reproducing the layout of the MEA) after the stimulation from site 46
—fourth column, sixth row. The small number reported in each box
represents the area of the histogram. Not responding channels are excluded. (b)
Responses evoked in the network by stimulation from site 62. As it can be
clearly notices the channels responding to site 46 respond also to channel 62,
denoting an absence of strong selection with respect to the stimulating
electrode. X-scale [ 0, 1 ]; Y-scale [ 0, 400 ] milliseconds.</p>
</caption>
<graphic xlink:href="CIN2007-12725.005"></graphic>
</fig>
<fig id="fig6" position="float"><label>Figure 6</label>
<caption><p>Connectivity maps (same data reported in <xref ref-type="fig" rid="fig4">Figure 4</xref>
).
The connectivity map represents a plot of the PSTH areas evoked by a couple of
stimulating electrodes on a specific electrode. In this way we are able to
represent the network response to a specific choice of stimulating sites. The
ideal case should be to have two recording electrodes placed on the axis, far
from the origin (i.e., maximum response to one stimulating electrode and zero
to the other). (a) Before the robotic experiment. (b) After the robotic
experiment.</p>
</caption>
<graphic xlink:href="CIN2007-12725.006"></graphic>
</fig>
<fig id="fig7" position="float"><label>Figure 7</label>
<caption><p>PSTH and STS in a neurorobotic experiment. (a) PSTHs
for two electrodes (chosen as recording —motor electrodes) with respect
to two stimulating sites. Electrode 15 responds well to stimulation from electrode
16 and bad to stimulation from electrode 48; on the contrary electrode 45
responds well to 48 and bad to 16. This tendency is maintained and even
improved after the robotic experiment (b). The STS graphs before (c) and after
(d) the robotic experiment prove again the selectivity of the chosen electrodes
and the improvement in the performances (increased STS area).</p>
</caption>
<graphic xlink:href="CIN2007-12725.007"></graphic>
</fig>
<fig id="fig8" position="float"><label>Figure 8</label>
<caption><p>Robot trajectories and performances in a neurorobotic
experiment. (a) Robot trajectory during the first free running phase. (b) Robot
trajectory during the last free running phase (i.e., after the learning phase).
(c) Indicators of the robot's performance. The last two parameters only show
that the two phases are comparable in terms of covered space and trajectory
length during the robot's movement inside the arena. For this reason, the
reduction of hits in the second phase (i.e., first parameter) suggests an
improvement of performances during the obstacle-avoidance task. The conclusion
is that an improvement in the robot's behavior in terms of a decreased number
of hits must depend from the modulation of the neuronal activity, as also
confirmed by the graphs of the STS presented in the previous figure.</p>
</caption>
<graphic xlink:href="CIN2007-12725.008"></graphic>
</fig>
</floats-wrap>
</pmc>
</record>
Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Ticri/CIDE/explor/TelematiV1/Data/Pmc/Corpus
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000442 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Pmc/Corpus/biblio.hfd -nk 000442 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien |wiki= Ticri/CIDE |area= TelematiV1 |flux= Pmc |étape= Corpus |type= RBID |clé= |texte= }}
This area was generated with Dilib version V0.6.31. |