Whole-Brain Connectomic Graph Neural Network Enables Whole-Body Locomotion Control in Drosophila

Zehao Jin, Yaoye Zhu, Chen Zhang, Yanan Sui

Keywords: Connectomics, Graph Neural Network, Embodied Intelligence, Sensorimotor Control

TL;DR: We show that a controller instantiated directly from the Drosophila whole-brain connectome can achieve diverse embodied locomotion behaviors, demonstrating connectome topology as a powerful inductive bias for deep reinforcement learning.

Abstract: Whole-brain connectome provides a structural blueprint for linking neural circuits to behavior, yet its application to embodied control remains largely unexplored. We introduce the fly-connectomic Graph Neural Network (flyGNN), a reinforcement learning controller whose architecture is instantiated directly from a complete adult Drosophila connectome. Our flyGNN models the connectome as a directed message-passing graph, partitioned into afferent, intrinsic, and efferent pathways that structure information flow from sensory inputs to motor outputs. Integrated with a dynamically controllable biomechanical model of Drosophila, flyGNN achieves stable control across diverse locomotion tasks, including gait initiation, walking, turning, and flight, without task-specific architectural tuning. These results demonstrate that whole-brain connectivity can directly support embodied reinforcement learning, establishing a new paradigm for connectome-based control algorithms.

Primary Area: applications to robotics, autonomy, planning

Submission Number: 11937