Using the soft actor-critic algorithm to generate rapid eye movements with an unconstrained 6DOF biomimetic robotic...

Understanding how the brain controls rapid eye movements, known as saccades, remains an open challenge. Biomimetic models have provided valuable insight into these control mechanisms. In this work, we develop an accurate, six-degree-of-freedom computational model of a biomimetic eye to explore the neural control signals underlying these movements. We hypothesise that saccade generation is governed by an optimisation of multiple costs, including accuracy, duration, energy, and tendon tension. To test this hypothesis, we trained a model-free reinforcement learning algorithm using biologically inspired input signals, in an open-loop manner. Our results show that the emerging control strategies successfully replicate human-like saccadic characteristics, including nonlinear main sequence relationships, compliance with Listing's law, straight trajectories, normometric pulse-step-like controls, and the antagonistic pairing of extraocular muscles, without explicitly enforcing these behaviours. Additionally, we analyse the impact of different reward costs and noise components on the resulting saccadic control strategies and resulting motions.