What it measures
Generalizable manipulation and locomotion skills in a fast, GPU-parallel simulator, with task success measured per environment. The paper emphasizes simulation throughput alongside task performance.
SimulationActiveMedium relevance
ManiSkill3
ManiSkill3: GPU Parallelized Robotics Simulation and Rendering for Generalizable Embodied AI
ManiSkill3 is an open-source, GPU-parallelized robotics simulator and benchmark built on the SAPIEN engine for generalizable embodied AI. The paper describes environments spanning 12 domains, including mobile manipulation and humanoids, and reports throughput of up to 30,000+ FPS in benchmarked environments. The robot library includes Unitree G1 and H1 models.
- First published
- 2024
- Organizers
- Stone Tao et al.
- Tasks
- —
- Simulator
- SAPIEN (GPU-parallel)
- License
- Apache-2.0 (code); CC BY-NC 4.0 (assets)
How it works
Environments run many copies in parallel on the GPU, with state-based or rendered visual observations, so reinforcement learning and imitation learning can be trained at high throughput. Success is defined per task environment. The paper reports rendering that is 10 to 1000 times faster and uses 2 to 3 times less GPU memory than other platforms, as measured on its benchmarks.
Why it matters
A fast simulator with humanoid models in its robot library makes large-scale reinforcement learning on humanoid bodies practical. The documentation opened lists the Unitree G1 and H1 models but does not say which tasks use them.
Scoring
Metrics
| Metric | Key | Unit | Direction |
|---|---|---|---|
| Task success rate (per environment) | success_rate | Unitless | Higher is better ↑ |
Hardware
Robots involved
The ManiSkill documentation lists Unitree G1 and H1 models (including simplified leg and upper-body variants) among supported robots. Most listed robots are arms and mobile manipulators. Which tasks use the humanoid models was not confirmed.
Recorded outcomes
Results
No result files yet
This record describes the evaluation, but no verified structured results have been added.
Evidence
Sources
- [1]ManiSkill3: GPU Parallelized Robotics Simulation and Rendering for Generalizable Embodied AI ↗arXiv · accessed Oct 8, 2026
- [2]ManiSkill (GitHub repository) ↗GitHub · accessed Oct 8, 2026
- [3]ManiSkill documentation: robots ↗ManiSkill project · accessed Oct 8, 2026
Related evaluations