A Biophysically Detailed C. elegans Circuit as a Task-Agnostic Dynamical Core for Visually Robust Robot Manipulation

2026-09-30

AI Summary

This study addresses the limited generalization of robotic policies across tasks, embodiments, and visual environments by proposing a visuomotor policy grounded in the biophysical neural circuits of *C. elegans* as its dynamical core. Methodologically, it employs multi-compartment neuron modeling and biophysical simulation, keeping evolutionarily derived synaptic weights fixed while training only lightweight task adapters. We demonstrate that visual robustness can be directly inherited from the intrinsic dynamics of these biophysical circuits, eliminating the need to learn task-specific controllers. Experiments show that this approach outperforms baseline models across multiple MetaWorld tasks and on a physical robotic manipulator, significantly mitigating performance degradation under visual perturbations.

Abstract

Robot policies are usually trained for one task, one body and one visual environment, and generalize poorly beyond these conditions. Whether a nervous system can instead supply the sensorimotor computation through its evolved wiring and biophysics remains unresolved. Here we embed a biophysically detailed Caenorhabditis elegans sensorimotor circuit - 136 multicompartment neurons with realistic morphologies and electrophysiological characteristics - as the dynamical core of a visuomotor policy. Only thin task-specific adapters are trained; the core's synaptic weights stay fixed while its membrane voltages evolve freely. Across different MetaWorld tasks the core matches or exceeds diffusion-policy, action-chunking-transformer and neural-circuit-policy baselines, and degrades less under visual perturbations. Replacing the core with generic network models such as MLP, LSTM, transformer or reservoir networks removes the advantage. Furthermore, on a real robotic arm the core withstands diverse visual perturbations that collapse the baselines. Our results suggest that visual robustness can be inherited from biophysically detailed circuit dynamics rather than learned by task-specific controllers.

Problem

• robot manipulation
• visual robustness
• generalization
• biophysical neural circuit
• sensorimotor computation

Innovation

• biophysically detailed circuit
• task-agnostic dynamical core
• visual robustness
• robot manipulation
• C. elegans