LeCAR-Lab/ASAP
RSS 2025 paper on using reinforcement learning to train humanoid robots agile whole-body skills with aligned simulation-to-real-world physics.
Not currently ranked — collecting fresh signals.
star history
The project implements a reinforcement learning pipeline for training humanoid robots to perform agile whole-body motor skills. It leverages NVIDIA’s physics simulation platforms (IsaacGym, IsaacSim, Genesis) and the HumanoidVerse framework to train motion-tracking skills with a delta action model. The core contribution is methods to align simulation physics with real-world dynamics to enable successful sim-to-real transfer of learned behaviors.
Frequently asked
- What is LeCAR-Lab/ASAP?
- RSS 2025 paper on using reinforcement learning to train humanoid robots agile whole-body skills with aligned simulation-to-real-world physics.
- Is ASAP open source?
- Yes — LeCAR-Lab/ASAP is open source, released under the MIT license.
- What language is ASAP written in?
- LeCAR-Lab/ASAP is primarily written in Python.
- How popular is ASAP?
- LeCAR-Lab/ASAP has 2.1k stars on GitHub.
- Where can I find ASAP?
- LeCAR-Lab/ASAP is on GitHub at https://github.com/LeCAR-Lab/ASAP.