Decentralized neural dynamics and sensory constraints shape brittle star locomotion

Abstract

Abstract Brittle stars move with remarkable whole-body coordination despite lacking a central brain. With five identical arms radiating from a central disc, they predominantly adopt a bilaterally symmetrical rowing gait: one arm leads, two neighbouring arms row in synchrony and the remaining arms trail. This raises a puzzle: how does a brainless nervous system generate coherent whole-body gaits, and why does it favour rowing? To address this, this work introduces an in silico framework combining (i) a three-dimensional model of brittle star morphology in a physics simulator, (ii) an artificial neural network (ANN) architecture that mirrors the decentralized arm-level ganglia, which interconnect through the nerve ring and (iii) reinforcement learning (RL) to optimize controllers for locomotion. Analysis of optimized controllers shows that ganglia behave as distributed oscillators whose coupling via the nerve ring yields synchronization, analogous to that of central pattern generators (CPGs). Gait analysis reveals that rowing emerges as the strategy most compatible with the arms’ dual role as effectors and sensors. Taken together, these results provide a mechanistic view of how decentralized neural dynamics and sensory constraints shape brittle star locomotion. The presented framework offers an open-ended test bed for hypotheses inaccessible in vivo, and more broadly, for exploring decentralized control in embodied agents. We provide a link to our project web page and interactive results dashboard: https://airo.ugent.be/projects/brittle-star.