How does the brain control the eye movements ? An analysis-by-synthesis approach

Despite extensive research on eye movements, the underlying mechanisms by which the brain controls these motions are not yet well understood. Biomimetic models of biological systems have provided valuable insights into their function and control. This work aims to develop an artificial model of the human eye system to explore how the brain controls saccadic movements.

It is hypothesised that the control of saccadic movement arises from an optimisation process to the control inputs, where metrics such as accuracy, duration, energy, and tension are used to evaluate the performance of the resulting motion. To this end, a model-free algorithm based on reinforcement learning is employed to learn, through trial-and-error, how to control the system in an open-loop manner, using biologically inspired input signals.

A biologically accurate, six-degree-of-freedom computational model of the human eye is used to analyse how different inputs affect the resulting ocular motion. The results derived from this formulation show that the learnt controls successfully replicate human-like saccadic characteristics, including the main sequence relationships, the compliance with Listing's law, and the antagonist pairing of extraocular muscles - without directly enforcing these behaviours.

Additionally, this formulation was used to analyse the impact of the reward function factors, as well as the introduction of signal-dependent and additive noise, on the resulting saccadic control strategies and resulting motions.