Biomemetic Robots by Dr. Joseph Ayers professor of biology at Northeastern University
Presentation in Cambridge Massachusetts on June 2, 2008
We are developing biomimetic robots based on neurobiological model systems, the lobster
and the lamprey. These robots feature:
A biomorphic plant engineered to capture the biomechanical advantages of the animal
model A neuronal network based controller.
Myomorphic actuators that feature graded control.
Neuromorphic Sensors that employ a labeled line code and A behavioral library based on
command sequences reverse engineered from movies of the animal model.
Existing implementations of these robots are based on finite state machine based controllers
that instantiate a set of finite state machines based on the organizational principles of the
animal model nervous systems. These state machines include leg or body axis central
pattern generators (CPGs) that generate leg movements or undulations, postural pattern
generators that control compensatory appendages and/or adaptive sensors and sensory
integration networks that process sensor information. A hierarchical command and
coordination level that sequences behavior makes choices and establishes the
intersegmental coordination pattern of the locomotory CPGs modulates the state machines.
Although these systems are quite effective in laboratory conditions their effectiveness in
unpredictable field environments degrades. We are exploring two types of electronic neuron
models in our controllers.
Hindmarsh-Rose (HR) electronic neurons are analog computers that simulate the Hindmarsh
Rose equations in real time. Discrete Time Map-based (DTM) neurons are computationally
efficient models based on a simple iterative map. Chemical-based synapse models are
present for both types of models and allow the generation of realistic neuronal circuits. The
key feature of these models is that because they are based on capturing of nonlinear
dynamical behavior of neurons rather than neuronal conductance models, they are simpler,
can operate in real time and are thus suitable for robotic control applications.
The ENS systems can be interfaced directly to actuators by using the EN action potentials to
gate power transistors supplying current to SMA actuators. We have implemented both HR
and DTM CPGs for the Lobster robot.
The HR CPG has been used to control a SMA based walking leg to walk in different
directions. By directing proprioceptive sensor feedback to the neuronal oscillator elements
of the CPG, we have been able to instantiate proprioceptive leg reflexes for load compensation
as well as reversing reflexes for stumbling and limb contact.
We have performed detailed modeling of the exteroceptive control of rheotaxic behavior
mediated by antennal sensors. Surge and flow is detected by strain gauges embedded in the
antenna that are bent by flow.
The sensory input is processed by two layers of interneurons that distinguish lateral versus
axial flow and activate walking command systems to mediate rotational and yawing
components of rheotaxis. A separate optical flow layer can fuse with this hydrodynamic flow
to provide more robust modulation.
We are also developing an ENS for the control of swimming in the lamprey robot. This
controller uses concatenated segmental CPGs to generate propagating axial flexions. A
layered control system includes, CPGs, recruiters, coordinating elements, command
elements, modulatory interneurons as well as an inclinometer-based “vestibular” system for
the control of pitch and roll.
Joseph Ayers is a professor of biology at Northeastern University.
Grey Thumb is inspired by the Homebrew Computer Club and by Make magazine, but is
geared mostly toward artificial life, agent based simulation, and evolutionary computation.
Our goal is to bring together people from diverse backgrounds to help create a nucleus of
interest in these subjects in the Boston area. (http://www.greythumb.org)
The Robots are Coming
Next
Presentation in Cambridge Massachusetts on June 2, 2008
We are developing biomimetic robots based on neurobiological model systems, the lobster
and the lamprey. These robots feature:
A biomorphic plant engineered to capture the biomechanical advantages of the animal
model A neuronal network based controller.
Myomorphic actuators that feature graded control.
Neuromorphic Sensors that employ a labeled line code and A behavioral library based on
command sequences reverse engineered from movies of the animal model.
Existing implementations of these robots are based on finite state machine based controllers
that instantiate a set of finite state machines based on the organizational principles of the
animal model nervous systems. These state machines include leg or body axis central
pattern generators (CPGs) that generate leg movements or undulations, postural pattern
generators that control compensatory appendages and/or adaptive sensors and sensory
integration networks that process sensor information. A hierarchical command and
coordination level that sequences behavior makes choices and establishes the
intersegmental coordination pattern of the locomotory CPGs modulates the state machines.
Although these systems are quite effective in laboratory conditions their effectiveness in
unpredictable field environments degrades. We are exploring two types of electronic neuron
models in our controllers.
Hindmarsh-Rose (HR) electronic neurons are analog computers that simulate the Hindmarsh
Rose equations in real time. Discrete Time Map-based (DTM) neurons are computationally
efficient models based on a simple iterative map. Chemical-based synapse models are
present for both types of models and allow the generation of realistic neuronal circuits. The
key feature of these models is that because they are based on capturing of nonlinear
dynamical behavior of neurons rather than neuronal conductance models, they are simpler,
can operate in real time and are thus suitable for robotic control applications.
The ENS systems can be interfaced directly to actuators by using the EN action potentials to
gate power transistors supplying current to SMA actuators. We have implemented both HR
and DTM CPGs for the Lobster robot.
The HR CPG has been used to control a SMA based walking leg to walk in different
directions. By directing proprioceptive sensor feedback to the neuronal oscillator elements
of the CPG, we have been able to instantiate proprioceptive leg reflexes for load compensation
as well as reversing reflexes for stumbling and limb contact.
We have performed detailed modeling of the exteroceptive control of rheotaxic behavior
mediated by antennal sensors. Surge and flow is detected by strain gauges embedded in the
antenna that are bent by flow.
The sensory input is processed by two layers of interneurons that distinguish lateral versus
axial flow and activate walking command systems to mediate rotational and yawing
components of rheotaxis. A separate optical flow layer can fuse with this hydrodynamic flow
to provide more robust modulation.
We are also developing an ENS for the control of swimming in the lamprey robot. This
controller uses concatenated segmental CPGs to generate propagating axial flexions. A
layered control system includes, CPGs, recruiters, coordinating elements, command
elements, modulatory interneurons as well as an inclinometer-based “vestibular” system for
the control of pitch and roll.
Joseph Ayers is a professor of biology at Northeastern University.
Grey Thumb is inspired by the Homebrew Computer Club and by Make magazine, but is
geared mostly toward artificial life, agent based simulation, and evolutionary computation.
Our goal is to bring together people from diverse backgrounds to help create a nucleus of
interest in these subjects in the Boston area. (http://www.greythumb.org)
The Robots are Coming
Next
 | |  | |  | |  |