ID 2369: Incremental Learning for motor control in neuromusculoskeletal models
The exact function of the cerebellum during motor learning is a topic of ongoing research. One hypothesis describes the role of the cerebellum as a forward model whose main functions are sensorimotor prediction and error processing.
It generates a prediction of the intended movement outcome based on a stored internal representation, which is then compared to the actual movement outcome. The resulting error leads to a signal that is sent back to the motor cortical and subcortical areas, which activates feedback movement corrections and calibration of the forward model.
Previous work has shown that the position of the next foot placement can be predicted from the center of mass state. The goal of this work is to implement this prediction model as biologically inspired neural network, which updates its predictions during the simulation, using incremental learning.
Tasks
- Implementation of a (simple) neural network that predicts the position of the next foot placement during the simulation (in SCONE).
- Implementation of an incremental learning approach to update the network and its predictions based on the error between the predicted foot placement and the actual foot placement.
Requirements
- Knowlegde in deep learning / neural networks / pytorch
- Interest in motor control / biologically inspired systems
