Active inference, eye movements and oculomotor delays

Area of Science:

• Computational neuroscience
• Robotics
• Control theory

Background:

• Sensorimotor delays pose challenges for optimal control in biological and artificial systems.
• Visuo-oculomotor loops are susceptible to temporal discrepancies between perception and action.
• Active inference offers a framework for understanding and implementing Bayes-optimal control under uncertainty.

Purpose of the Study:

• To investigate sensorimotor delays in smooth eye movements using active inference.
• To develop a computational model that compensates for delays in the visuo-oculomotor system.
• To explore the neurobiological plausibility of active inference for real-time control.

Main Methods:

• Utilized a generalized Kalman filtering approach within an active inference framework.
• Employed neuronal simulations to model pursuit initiation responses with and without delay compensation.
• Extended the generative model to simulate smooth pursuit eye movements and hierarchical trajectory recognition.

Main Results:

• Demonstrated that representing hidden states in generalized coordinates effectively compensates for sensory and oculomotor delays.
• Simulations showed successful compensation in pursuit initiation responses.
• The hierarchical generative model enabled recognition of occluded trajectories and anticipatory responses.

Conclusions:

• Active inference provides a unified and neurobiologically plausible solution for integrating information with temporal delays.
• The proposed method addresses challenges in controlling systems with delayed feedback, such as the oculomotor system.
• This framework supports anticipatory control and robust navigation in dynamic environments.