Whole-Brain Connectomic Graph Model Enables Whole-Body Locomotion Control in Fruit Fly

Zehao Jin; Yaoye Zhu; Chen Zhang; Yanan Sui

cs.LGFeb 20, 2026

Whole-Brain Connectomic Graph Model Enables Whole-Body Locomotion Control in Fruit Fly

Authors: Zehao Jin, Yaoye Zhu, Chen Zhang, Yanan Sui

Organizations: Tsinghua University

Abstract

Animals perform coordinated whole-body movements under the control of neural systems shaped by brain-wide connectivity. The mapping of the whole-brain neural connections, or the connectomes, provides a natural graph for modeling sensorimotor information flow, yet its potential as a neural controller for embodied agents remains largely unexplored. Here, we introduce the Fly-connectomic Graph Model, which directly instantiates the whole-brain connectome of an adult Drosophila as a graph-structured neural controller for movements of a simulated biomechanical fruit fly via deep reinforcement learning. We achieve stable performance across diverse locomotion tasks, as well as better sample efficiency compared to both graph and non-graph baselines. Our results demonstrate a biologically informed way towards effective control policy design by translating whole-brain wiring principles into actionable architectural priors, while also improving the interpretability through dynamic information flow. This work also highlights the potential to bridge neuromechanics with embodied intelligence by providing a computational platform for investigating the sensorimotor transformation underlying animal behavior and a paradigm to advance the development of more nature-aligned intelligent systems.

Figures & tables

Figure 1: Overview of the FlyGM-enabled whole-body locomotion control framework. Observations are mapped to afferent neuron states via a lightweight input projection. Neural states are propagated through a connectome-constrained message-passing module, and the resulting efferent states are decoded into motor actions that drive the embodied Drosophila model in MuJoCo. Demonstration videos and source code are available at: https://sites.google.com/view/flygm .

Figure 2: Structure of the Fly-connectomic Graph. (a) Aggregated synapse graph of the fly connectome, grouped into afferent (blue), intrinsic (green), and efferent (orange) sets across left hemibrain, central, and right hemibrain compartments. Node sizes reflect the number of neurons in each group, and arrows indicate the direction and relative strength of connectivity. (b) Force-directed graph layout [ 17 ] of the same neural network. The spatial layout reveals hemispheric symmetry and functional clustering.

Figure 3: Imitation learning dynamics across topologies. (a) Total training loss and (b) mean squared error of action means ( $\mu$ MSE) relative to expert trajectories. Shaded areas denote standard deviation across multiple runs. ER-Random: Erdős-Rényi random graph; Rewired: degree-preserving rewired graph; MLP: multilayer perceptron.

Model	Metric	$v{=}2,\psi{=}0$	$v{=}3,\psi{=}0$	$v{=}3,\psi{=}4$	$v{=}3,\psi{=}7$
FlyGM (ours)	Angle Err $\downarrow$	$\mathbf{4.96{\pm}0.09}$	$\mathbf{5.57{\pm}0.06}$	$\mathbf{6.36{\pm}0.33}$	$\mathbf{8.29{\pm}0.21}$
DP Rewiring	Angle Err $\downarrow$	$7.84{\pm}0.08$	$7.77{\pm}0.08$	$9.72{\pm}0.12$	$13.55{\pm}0.69$
Erdős-Rényi	Angle Err $\downarrow$	$12.11{\pm}0.23$	$11.33{\pm}0.27$	$17.45{\pm}0.26$	$125.36{\pm}8.96$
MLP	Angle Err $\downarrow$	$6.76{\pm}0.13$	$7.18{\pm}0.05$	$8.85{\pm}0.24$	$13.90{\pm}0.45$

Table 1: Control performance across speed (

v

, cm/s) and yaw (

\psi

, rad/s) conditions after the full IL+PPO pipeline. Angle error in degrees. Values are mean

\pm

std with the per-column best mean in bold. DP Rewiring denotes the degree-preserving rewiring of the connectome. SNN baseline results are reported separately in Appendix F .

Panel of Figure 4 — Figure 4: Walking task snapshots. (a) Snapshots of the simulated fly during the gait initialization phase. (b) Snapshots of the simulated fly walking in a straight line at a velocity of 3 cm/s.

Panel of Figure 5 — Figure 5: Turning and flight task snapshots. (a) Snapshots of the virtual fly executing a high-speed left turn at 3 cm/s and 10 rad/s. (b) Snapshots of the virtual fly executing a straight flight task at a velocity of 20 cm/s.

Figure 6: Neural and behavioral dynamics across a continuous walking sequence. (a) Force-directed layout of the connectome at 200 ms, with neurons colored by flow type and representation intensity. (b) Behavioral snapshots (0-680 ms) show straight walking, decelerating, pausing, speeding-up, and turning phases; blue dots and ghost fly mark target trajectory and orientation. (c) Temporal dynamics across the connectome, grouped by flow type and major neuron classes and showing transitions between behavioral phases.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 7: Joint kinematics and actuator activations during walking. We visualize the angles (blue) and actuator activations (red) of coxa joints from left (T1-T3) and right (T1-T3) legs over multiple gait cycles. Dashed vertical lines mark gait phases based on the troughs of the left T1 coxa. The model reproduces alternating tripod-like coordination, with left T1/T3 synchronized with right T2, and left T2 synchronized with right T1/T3, consistent with expected fly walking patterns.

Figure 8: Neural and behavioral dynamics of the Erdős-Rényi random-graph control. Same analysis pipeline as Figure 6 , but applied to a policy whose underlying graph is an Erdős-Rényi random graph with the same number of nodes and edges as the connectome. Activation patterns are largely uniform across superclasses, with no clear functional differentiation between sensory, central, and motor groups, in contrast to the structured dynamics observed in FlyGM.

Metric	$v{=}2,\psi{=}0$	$v{=}3,\psi{=}0$	$v{=}3,\psi{=}4$	$v{=}3,\psi{=}7$
Angle Err $\downarrow$	$94.87{\pm}20.14$	$108.17{\pm}17.83$	$133.17{\pm}0.00$	$125.73{\pm}0.00$

Appendix

Table 2: SNN baseline control performance under the same conditions as Table 1 . Angle error in degrees. Under high-yaw conditions the policy terminates immediately at gait initiation, so all seeds collapse onto the same worst-case error and the std is essentially

0

FlyGM Variant	Metric	$v{=}2,\psi{=}0$	$v{=}3,\psi{=}0$	$v{=}3,\psi{=}4$	$v{=}3,\psi{=}7$
Weighted (full)	Angle Err $\downarrow$	$4.96{\pm}0.09$	$\mathbf{5.57{\pm}0.06}$	$6.36{\pm}0.33$	$\mathbf{8.29{\pm}0.21}$
Unweighted	Angle Err $\downarrow$	$4.91{\pm}0.13$	$5.90{\pm}0.10$	$6.93{\pm}0.12$	$11.00{\pm}0.27$
No Desc + Weighted	Angle Err $\downarrow$	$5.26{\pm}0.12$	$6.02{\pm}0.11$	$7.03{\pm}0.08$	$9.17{\pm}0.17$
No Desc + Unweighted	Angle Err $\downarrow$	$5.00{\pm}0.05$	$6.59{\pm}0.13$	$8.52{\pm}0.13$	$13.23{\pm}0.20$
Shared Desc + Weighted	Angle Err $\downarrow$	$\mathbf{4.25{\pm}0.21}$	$5.59{\pm}0.22$	$\mathbf{6.01{\pm}0.17}$	$9.10{\pm}0.38$
Shared Desc + Unweighted	Angle Err $\downarrow$	$4.75{\pm}0.09$	$6.61{\pm}0.08$	$8.65{\pm}0.16$	$13.43{\pm}0.67$

Appendix

Table 3: Ablation of edge weights and intrinsic descriptors in FlyGM, evaluated under the same conditions as Table 1 . Angle error in degrees. Values are mean

\pm

std with per-column best in bold.

Model	Dim	Depth	Eval KL $\downarrow$	Avg. Reward $\uparrow$
GCN	16	4	7.93	43.13
EdgeCNN	4	2	5.54	90.48
GAT	8	2	3.43	132.65
GraphSAGE	16	4	3.31	125.55
PNA	4	2	2.89	145.33
FlyGM (ours)	32	4	2.34	334.34

Appendix

Table 4: Comparison of FlyGM against standard GNN architectures on the walking task. Eval KL measures the KL divergence to the expert policy at the end of the imitation learning stage; Avg. Reward is the mean episodic return under the RL reward function after the full IL+PPO pipeline. Dim and Depth denote node embedding dimension and number of message-passing layers, respectively. For each baseline, these were set to the largest configuration that fits within a single A100 80 GB GPU.

Whole-Body Control

Whole-body control coordinates a robot’s limbs, torso, and mobility to perform tasks while maintaining balance, contact, and safety.

See all 59 papers

FlyCNS: Connectome-Grounded Information Organization for Communication-Constrained Embodied Control
Sep 23, 2026Jinchang Zhang, Jiakai Lin, Guoyu LuRobotic Control Robot Locomotion
Intelligent Vision and Sensing (IVS) Lab at Indiana University Bloomington
Sensorimotor behaviour from an untrained brain-and-nerve-cord connectome
Oct 24, 2026John R. Minnick, Jesus Gonzalez-Ferrer, Jinghui Geng +4Neural Processes Embodied Navigation
Department of Electrical and Computer Engineering, University of California, Santa Cruz, CA, USA. · UC Santa Cruz Genomics Institute, University of California, Santa Cruz, CA, USA. · Department of Biomolecular Engineering, University of California, Santa Cruz, CA, USA. +1
FLYNN: Robust Neural Network for Robot Navigation using Fly Brain Topology
Jun 21, 2026Benquan Wang, Jingdao ChenRobot Navigation Neural Network Robustness
Department of Computer Science Engineering, Mississippi State University, Mississippi State, MS 39762, USA

Sep 23, 2026J/K move · Enter open · S save

FlyCNS: Connectome-Grounded Information Organization for Communication-Constrained Embodied Control

Jinchang Zhang, Jiakai Lin, Guoyu Lu

Intelligent Vision and Sensing (IVS) Lab at Indiana University Bloomington

Robotic Control Robot Locomotion

Open paper AskPDF arXiv

Robotic bodies are inherently distributed in sensing and actuation, yet learning-based control still commonly relies on centralized information processing. This work studies the problem of information organization in communication-constrained embodied control: which computations should remain local, and which information is worth transmitting for whole-body coordination. We propose FlyCNS, an embodied information-organization framework inspired by the Drosophila brain--nerve-cord connectome. FlyCNS preserves local sensorimotor computation within each limb and enables selective long-range communication through separate ascending and descending routing pathways. From a real connectome, FlyCNS extracts the directional structural complexity of these two pathway types and uses it as a weak prior over communication allocation, while message content, transmission timing, and locomotion policies remain task-adaptive and are learned through reinforcement learning. In Unitree Go1 simulation, FlyCNS exhibits more graceful performance degradation as the communication budget is tightened. Under the most restrictive setting, it uses only about 21--22% of the communication of the full-communication reference, while still maintaining a tracking score of approximately 0.882 under both command protocols, with a gap of no more than 6.1% from the full-communication reference. These results indicate that real neural connectomes can inform not only the structural design of control networks, but also provide transferable inductive biases for information organization across embodiments, guiding robots in balancing local computation and long-range coordination under limited communication resources.

Abstract

Whole-Body Control

Explore similar work

FlyCNS: Connectome-Grounded Information Organization for Communication-Constrained Embodied Control

Sensorimotor behaviour from an untrained brain-and-nerve-cord connectome

FLYNN: Robust Neural Network for Robot Navigation using Fly Brain Topology