LAW
System that pairs a VLM, which handles language and acts as a value expert, with a world-action model, so that one loop covers language, action and world modeling. Checkpoints are released for LAW on Fast-WAM and on OpenWAM-α; the OpenWAM-α variant scores higher on RoboTwin 2.0 in the paper (95.56).
Peking University & Beijing Innovation Center of Humanoid Robotics · Oct 16, 2026 · Agent system · built on Fast-WAMReport an error
1Benchmarks reported
Oct 16, 2026Released
OpenWeights
Agent systemType
Benchmark results1
- 94.2%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Oct 2026· self-reported, 50 clean + 500 randomized demos· n = 10000· non-standard
94.2%, self-reported by the authors, not a leaderboard entrynot ranked
Not reported on 14 benchmarks
Wrong or missing number? Report an error