CHORUS trains one VLA to coordinate multiple robots on manipulation tasks

chris_j_paxton · x · 2026-07-29

The repost highlights CHORUS, a system that trains a single VLA to coordinate multiple diverse robots using only local observations and identifying information.

The key claim is scalability: instead of relying on bespoke centralized control, the same model can support collaboration on manipulation tasks such as measuring with a tape, handing over a book, and lifting a laundry basket.

Related event: CHORUS: Single VLA Model Enables Decentralized Multi-Robot Collaboration(6 posts)→

Original post →

More from Embodied

Embodied channel →