Goal-driven neural network models of biological motor control - a neurorobotic study

Biological agents display impressive abilities to move under dynamic environmental conditions. Understanding how the brain controls movements in biological systems can potentially help build better control systems for robots that are inspired by (or) built to mimic the biological system complexity. There are several important issues to be solved such as how complex temporal activity is generated by the central nervous system to control actuation (movement generation), how does the system cope with changing environment and body dynamics (movement adaptation), and the mechanisms behind the prediction of body states. Distributed regions in the central nervous system coordinate to produce the motor commands suitable for a given behavioural goal. Even more so, the same neural substrates that are involved in movement production are observed to be actively participating in adjusting the motor responses with respect to various sources of disturbances and uncertainties. However, the neural computations that underlie even the simplest of the movements are still equivocal. Historically, one of the main sources of the dispute has been due to the emphasis on describing the motor-encoding in isolated brain regions, by ignoring the effect of key control elements such as sensory-feedback, prediction and the physics of the body under control. In this dissertation, we demonstrate the utility of mechanistic neural networks as normative models to study neural computations underlying movement control. Particularly, we emulate recent observations from the primate motor cortex and the cerebellum - the brain regions that are implicated for generating motor-plans and predicting the sensory consequences of movement respectively. To arrive at a ground truth, the control simulations were confined to tasks such as point-to-point limb reaching movements, loaded postural perturbations, and ballistic eye-movements. Upon emulating neural activities in simulations, different robotic platforms, such as humanoids and soft-robots, were used to test the computational sufficiency of the inferred control and adaptation algorithms.