flybody simulates a fruit fly’s movements with open-source code, a user says
The post credits Google DeepMind and HHMI Janelia with the project and says an AI agent learns to move the simulated body rather than relying on animation.
TLDR
A user describes flybody as a Python-based fruit fly simulation rebuilt from microscopy and running in MuJoCo, a physics engine. The post says it includes articulated legs, wings, head and abdomen, plus leg adhesion to grip surfaces. According to the user, its reinforcement-learning environments cover walking imitation, free flight and vision-guided flight, where the agent uses what the simulated fly sees. The user says the code is available on GitHub under the Apache 2.0 license.
Combined views
566.7K
1 Source, first seen 19d ago