MAWAM improves multi-robot coordination success to 67.8%
November 20, 2026
MAWAM uses cross-agent attention to jointly generate future views and action chunks for multiple robots. It achieves a 67.8% mean success rate on real-robot tasks, outperforming MA-VLA by 18.9 percentage points.
HOW THIS AFFECTS YOU
●
builderThis could help you build robot teams that handle complex tasks like folding towels or handing over objects.
●
researcherYou can use cross-agent attention to coordinate multiple robot arms in a single sample.