CHORUS: Decentralized Multi-Embodiment Collaboration with One VLA Policy

CHORUS is a novel framework that enables decentralized multi-robot collaboration using a single vision-language-action (VLA) model backbone, allowing each robot to operate based solely on its local observations. In experiments, CHORUS demonstrated a 64% performance improvement over traditional decentralized models and a 40% increase in reactivity to teammate behavior, while also outperforming centralized approaches. This advancement is significant for practitioners as it simplifies the deployment of multi-robot systems by eliminating the need for individual policies or inter-robot communication during inference.