SJTU AutoLab proposes A²-MEAS: correction-code multi-agent framework for embodied AI
AutoLab-SJTU · hf · 2026-11-10
Shanghai Jiao Tong University's AutoLab and Anyverse Dynamics published Toward the Era of Agentic Embodied Intelligence, introducing A²-MEAS, an embodied multi-agent framework built on "code as a scaling interface."
- A correction code is a learned representation of corrective language, grounded spatial evidence, execution history, and an action proposal; MetaBrain supplies corrective supervision during training and grounded guidance at execution
- The correction encoder/adapter absorbs corrective knowledge into trainable policy weights; at deployment all weights stay frozen while fresh codes adjust behavior to handle new failures
- The underlying A²-WLA action model augments a VLA policy with motion-aware history and a latent world model, reportedly reaching 0.87555 success via action pretraining alone (arXiv, Mar 2026)
More from Embodied
- Caltech robot builds a structure in orbit autonomously, a first for space construction — GregCook2011 · 2026-11-10
- Day 5 of a DIY robot arm: CAD render of the differential wrist in progress — _Stocko_ · 2026-11-10
- SuperMimic's 30× gradient SNR edge over score-function cuts imagined steps 64× vs PPO — daibond_alpha · 2026-11-10
- One frozen controller for humanoids: spectral skills enable motion composition and language control — daibond_alpha · 2026-11-10
- SuperMimic trains humanoid tracker in 3 hours on one GPU, beating 128-GPU SoTA baseline — daibond_alpha · 2026-11-10
- Johns Hopkins lab hiring postdocs in embodied intelligence, applications due Dec 11 — mangahomanga · 2026-11-10