MIT researchers propose D-VLC for multi-robot collaboration
Decentralized AI framework lets robot swarms tackle unseen tasks 55% faster
D-VLC is a decentralized vision-language collaboration framework that lets heterogeneous robot swarms interpret complex instructions, decompose tasks, and coordinate in unknown environments without centralized control or task-specific training. By combining decentralized asynchronous reasoning, lightweight information sharing, capability-aware collaboration, and a unified action interface, it enables general-purpose VLMs to generate robot-specific actions executed by learning-free experts. Across diverse scenarios and multiple VLMs, the framework achieves success rates above 70% and cuts completion time by up to 55.8% relative to the geometric greedy baseline.
- D-VLC enables decentralized, asynchronous collaboration between heterogeneous robots using only VLMs and lightweight communication
- Achieves 70%+ success rates and 55.8% faster task completion than traditional methods in unknown environments
- Eliminates need for predefined task models or robot-specific training via a unified action interface
Why It Matters
This could enable faster deployment of robot swarms for disaster response, warehouse automation, and exploration without costly custom programming.