aiqbench analysis
CHORUS demonstrates that a single shared Vision-Language-Action (VLA) backbone can enable reactive, decentralized multi-robot collaboration using only local observations, eliminating the need for explicit inter-robot communication or per-robot policy alignment at inference time. Real-world experiments show significant performance gains over decentralized baselines and centralized methods, suggesting that pretrained visuomotor priors are sufficient to scale coordination across diverse robot teams without communication overhead. This approach addresses a key bottleneck in mobile robotics by leveraging existing VLA models to simplify deployment and improve scalability in partial observability scenarios.
CHORUS: Decentralized Multi-Embodiment Collaboration with One VLA Policy
arXiv:2606.12352v1 Announce Type: new Abstract: Multi-robot collaboration allows robots to efficiently take on a wide range of tasks, from moving a couch through a doorway to assembling structures on a construction site. However, achieving such coordination in mobile multi-robot settings remains challenging: centralized methods conditioned on the combined observations of a team scale poorly with team size, and decentralized methods that train one policy per robot often require explicit alignment procedures or information sharing at inference time to overcome partial observability. Our key insight is that the visuomotor priors of pretrained vision-language-action (VLA) models should enable reactive, decentralized collaboration from each robot's local observations alone, without these inference-time assumptions. We propose CHORUS, a framework that adapts a single VLA backbone to control diverse, multi-robot teams. At inference time, each robot runs an independent copy of CHORUS, conditioned only on its own observations and a robot-identifying prompt. In real-world experiments including mobile tape measurement, library book handovers, and laundry basket lifting, CHORUS achieves a 64% point improvement over decentralized, from-scratch models, improves reactivity to teammate behavior by 40% points, and outperforms centralized baselines. Together, these results show that a shared VLA backbone is capable of achieving decentralized multi-robot collaboration, without per-robot policies or inter-robot communication at inference.
Abstract:Multi-robot collaboration allows robots to efficiently take on a wide range of tasks, from moving a couch through a doorway to assembling structures on a construction site. However, achieving such coordination in mobile multi-robot settings remains challenging: centralized methods conditioned on the combined observations of a team scale poorly with team size, and decentralized methods that train one policy per robot often require explicit alignment procedures or information sharing at inference time to overcome partial observability. Our key insight is that the visuomotor priors of pretrained vision-language-action (VLA) models should enable reactive, decentralized collaboration from each robot's local observations alone, without these inference-time assumptions. We propose CHORUS, a framework that adapts a single VLA backbone to control diverse, multi-robot teams. At inference time, each robot runs an independent copy of CHORUS, conditioned only on its own observations and a robot-identifying prompt. In real-world experiments including mobile tape measurement, library book handovers, and laundry basket lifting, CHORUS achieves a 64% point improvement over decentralized, from-scratch models, improves reactivity to teammate behavior by 40% points, and outperforms centralized baselines. Together, these results show that a shared VLA backbone is capable of achieving decentralized multi-robot collaboration, without per-robot policies or inter-robot communication at inference.
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2606.12352 [cs.RO]
(or arXiv:2606.12352v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.12352 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 297 words
Related coverage
MASkillBlender: Decentralized Whole-Body Coordination for Multi-Humanoid Loco-Manipulation via Skill Blending
arXiv · Oct 2, 2026
SynAgent: Generalizable Cooperative Humanoid Manipulation via Solo-to-Cooperative Agent Synergy
arXiv · Aug 4, 2026
Learning Self-Orchestration in Decentralized Agent Swarms
arXiv · Oct 17, 2026
CoRE: Learning Collaboration-Role Experts for Decentralized Collaborative Manipulation with One Policy
arXiv · Oct 7, 2026