A brain-machine system for neurobiology

Neuro-robotic systems provide a unique paradigm for relating computational questions to observations of neuronal behavior. We describe a new experimental preparation for studying the neurobiological underpinnings of motor learning through the interaction of a mobile robot with live neural tissue. The brainstem of a lamprey was stimulated by electrical impulses that coded the light intensity detected by sensors on the robot. The neural responses to these stimuli controlled the speed of the robot's wheels. In this closed, loop arrangement, the robot responded to a light source with a behavior that reflected the neural processing in the brainstem. This neuro-robotic system allowed us to test specific hypotheses on neural information processing based on the observed behavior of the robot and on the recorded neural activities. We compared the performance of the actual system with the simulated performance obtained from different neural network models. We found that dynamic networks with recurrent dynamics are significantly superior to static feedforward networks model, even when the dynamic models have fewer parameters than the static models. Additional findings led us to conclude that the main origin of this dynamic behavior is local ipsilateral influence of the previous state on the current state.