You are here: Home BCCN Research C4: A neuronal model of adaptive movement control
Document Actions

C4: A neuronal model of adaptive movement control

Carsten MehringJ, Stefan RotterB, Marc-Oliver GewaltigD and Klaus VogtJ

J = Animal Physiol. & Neurobiol, Inst. of Biol.I
B = Bernstein Center for Computational Neuroscience
D = Honda Research Institute Europe GmbH

Scientific background

A conceptual break-through in unraveling the dynamics of cortical motor control was achieved recently by demonstrating how real-time brain-control of neuroprosthetic effectors can be learned under closed-loop conditions. Activity of a small population of motor cortical neurons was used to decode intended 3D movements. The tuning of neurons changed when used for brain-controlled movements, improving the accuracy of the read-out. There is an ongoing debate what is represented by neural activity: direct muscle activations or abstract kinematic variables like the tangential velocity of the hand, or compound movement patterns. In a model-based joint analysis of physiological and behavioral data we found that single neurons can represent multiple movement variables simultaneously, modulated by contextual information. Another recent study revealed that decoding movements from brain activity can also be based on local field potentials that represent neuronal activation lumped over a certain volume. This indicates that distributed networks of yet unknown functional architecture, rather than individual neurons or small neuronal modules, are underlying movement control and motor learning.


Our objective is to analyze the control task underlying the acquisition and adaptation of voluntary arm movements, and to work out its underlying neuronal mechanisms. In particular, we want to elucidate the role of proprioceptive and reward-mediating feedback, the nature of variability both on the behavioral and the neuronal level, and the interplay of synaptic plasticity6 and functional adaptivity at the system level. Building on behavioral and physiological data obtained in collaborative projects (e.g. C5), we will devise and examine model systems for adaptive movement control that share essential structural and dynamic features with the biological system. We will finally compare our neuronal models with known engineering solutions and explore technical applications to task related learning and behavioral control for a humanoid robot system (Honda ASIMO).

Personal tools