chengxuxin/extreme-parkour

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1161 stars · 179 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official code for 'Extreme Parkour with Legged Robots' (ICRA 2024), a reinforcement learning framework for training quadruped robots to perform parkour using Isaac Gym simulation. It trains a single neural network policy end-to-end from depth camera images to precise locomotion control, deployable to low-cost hardware in under 20 hours of training.

Use cases

• train a quadruped robot to climb and jump obstacles
• simulate legged robot locomotion with reinforcement learning
• distill a privileged policy into a camera-based vision policy
• deploy RL locomotion policies to real robots
• reproduce parkour robot research results
• train robot policies on a single GPU

When to choose

• you want to train vision-based legged locomotion policies in simulation
• you have an NVIDIA GPU and Isaac Gym and want a proven sim-to-real quadruped pipeline
• you are reproducing or extending the Extreme Parkour paper

When to avoid

• you need a maintained production robotics framework with support and documentation
• you don't have access to NVIDIA Isaac Gym or a CUDA GPU
• you work with non-quadruped robots or manipulation tasks

Facets

library · maturity active

reinforcement-learning simulation machine-learning robotics robotics reinforcement-learning simulation machine-learning python legged-robots isaac-gym sim-to-real quadruped locomotion research-code parkour linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="chengxuxin/extreme-parkour")