Whole-body simulation of realistic fruit fly locomotion with deep reinforcement learning

janelia7_blocks-janelia7_biblio_header | block

Nature. 2025 Apr 23;. doi: 10.1038/s41586-025-09029-4

Whole-body simulation of realistic fruit fly locomotion with deep reinforcement learning Card LabReiser LabBranson LabTuraga LabProject Technical Resources

Vaxenburg Roman, Siwanowicz Igor, Merel Josh, Robie Alice, Morrow Carmen, Novati Guido, Stefanidi Zinovia, Card Gwyneth, Reiser Michael, Botvinick Matthew, Branson Kristin, Tassa Yuval, Turaga Srinivas

janelia7_blocks-janelia7_biblio_abstract | block

Abstract

The body of an animal influences how its nervous system generates behavior1. Accurately modeling the neural control of sensorimotor behavior requires an anatomically detailed biomechanical representation of the body. Here, we introduce a whole-body model of the fruit fly Drosophila melanogaster in a physics simulator. Designed as a general-purpose framework, our model enables the simulation of diverse fly behaviors, including both terrestrial and aerial locomotion. We validate its versatility by replicating realistic walking and flight behaviors. To support these behaviors, we develop new phenomenological models for fluid and adhesion forces. Using data-driven, end-to-end reinforcement learning we train neural network controllers capable of generating naturalistic locomotion along complex trajectories in response to high-level steering commands. Additionally, we show the use of visual sensors and hierarchical motor control, training a high-level controller to reuse a pre-trained low-level flight controller to perform visually guided flight tasks. Our model serves as an open-source platform for studying the neural control of sensorimotor behavior in an embodied context.

Preprint: www.biorxiv.org/content/early/2024/03/14/2024.03.11.584515