Overview
VNL-Playground is a fast, biologically realistic virtual environment for simulating animal behavior. It provides a platform for training virtual rodents with biomechanically realistic body models using deep reinforcement learning.
Built on top of MuJoCo-MJX and Ray for distributed training, it enables vision-guided, task-driven RL for virtual rodents at scale.
Key features
- Biomechanically realistic rodent body models
- GPU-accelerated physics via MuJoCo-MJX
- Distributed training via Ray
- Task-driven and vision-guided RL environments
Related work
This project is the foundation of the VNL-Ray master’s thesis: Biomechanically Realistic Virtual Rodent Behavior and Task Learning via Deep Reinforcement Learning