Reinforcement Learning on a rust simulation that only gives delayed rewards.
-
Updated
Apr 29, 2026 - Python
Reinforcement Learning on a rust simulation that only gives delayed rewards.
NVIDIA Isaac Lab / Isaac Sim task: train Unitree G1 humanoid to run and jump obstacle-course hurdles. Hierarchical RL with AMP locomotion, frozen BeyondMimic motion tracker (nepher-ai/humanoid-g1-tracking), and PPO high-level run/jump switcher. Clips from Hugging Face bones-studio/seed. skrl. Gym: Nepher-G1-Run, Jump, RunJumpHL.
Staged, reproducible RL hyperparameter search for Isaac Lab + skrl: declare a task in one YAML, run screen → refine → verify with fair canonical evals, live early-stop of dead runs, shared-GPU guards and termination-cause reports. A winner only counts after it reproduces.
NVIDIA Isaac Lab / Isaac Sim task for Unitree G1 humanoid RL: whole-body motion tracking, hierarchical AMP locomotion (skrl), and waypoint racing. Race PPO (10 Hz) commands a frozen AMP actor over a frozen tracker (50 Hz). Train with RSL-RL; EnvHub-reproducible race eval. Nepher Robotics.
Gridworld problem with dynamic obstacles implemented in PyGame GUI and solved using traditional RL agent (Q-learning) and deep RL agent (PPO algorithm).
Add a description, image, and links to the skrl topic page so that developers can more easily learn about it.
To associate your repository with the skrl topic, visit your repo's landing page and select "manage topics."