DECOWAM Improves Legged Robot Action Prediction by 21.7%
August 21, 2026
DECOWAM decouples camera ego-motion from base and arm actions in whole-body world-action models for mobile manipulation. By using residual adapters and adversarial separation of base and arm latents, the model reduces action MSE by 21.7% compared to the FastWAM backbone.
HOW THIS AFFECTS YOU
●
builderYou can use this approach to improve prediction accuracy in complex, multi-joint robotic systems.
●
researcherThe method offers a way to disentangle motion latents in world models.