Researchers' EFLUX uses LLMs to navigate multi-robot formations through cluttered spaces

Multi-robot teams operating in confined or cluttered environments must adapt both their formation geometry and group topology to navigate complex obstacles. Existing methods often model deformation (reshaping while staying connected) and reconfiguration (splitting into subgroups or merging) independently, using handcrafted rules that can lead to deadlock and suboptimal trajectories. EFLUX, developed by Jinyuan Zhang, Yuwei Wu, and colleagues, addresses this by extracting a structured scene representation and using an LLM to reason jointly over deformation actions like scaling and shearing, and reconfiguration actions like splitting and merging. The LLM-generated strategies are translated into executable per-robot waypoints through a closed-loop generation, verification, and correction pipeline, ensuring safe and continuous movement.

Simulation and hardware experiments demonstrate that EFLUX enables elastic formation navigation in constrained environments, significantly reducing deadlock and navigation failures compared to baseline methods. The framework maintains coherent multi-robot coordination while adapting to obstacles in real time. This work represents a step toward more intelligent, autonomous multi-robot systems that can handle dynamic scenarios without rigid pre-programmed rules, with potential applications in search and rescue, warehouse logistics, and exploration.