EFLUX: Elastic Multi-Robot Formation Navigation and Adaptation with Agentic LLMs @KumarLabPenn
EFLUX: Elastic Multi-Robot Formation Navigation and Adaptation with Agentic LLMs  @KumarLabPenn
Uploaded November 2026 | Updated November 2026, 1 day ago
In this work, we propose EFLUX, a context-aware, geometry-grounded LLM agentic framework for automatic and elastic multi-robot formation navigation. EFLUX extracts a structured scene representation and uses an LLM to reason jointly over both deformation actions (e.g., scaling and shearing) and reconfiguration actions (e.g., splitting and merging). These strategies are then translated into executable per-robot waypoints through a closed-loop generation, verification, and correction pipeline. Simulation and hardware experiments show that EFLUX enables safe, continuous, and elastic formation navigation in challenging scenarios, reducing deadlocks and failures while maintaining coherent multi-robot coordination.

Check our paper for more details: arxiv.org/abs/2607.12050
EFLUX: Elastic Multi-Robot Formation Navigation and Adaptation with Agentic LLMsControl of formation shape and position/orientationThe Picobug : a mesoscale robot that can run, fly, and graspRuzena Bajcsys NAE Simon Ramos Founder Award Acceptance SpeechDARPA Fast Lightweight Autonomy - UPenn Phase #2 FlightTrajectory Generation and Control for a Quadrotor with a Cable-Suspended PayloadVIO-Swarm: A Swarm of 250 g Vision Based QuadrotorsSearch-based Motion Planning for Aggressive Flight in SE(3)ModQuad: The Flying Modular Structure that Self-Assembles in MidairDeep Learning for Optimization of Trajectories for QuadrotorsInformation Theoretic Exploration in Signed Distance FieldsCooperative Localization and Mapping of MAVs using RGB-D Sensors
Vijay Kumar |

EFLUX: Elastic Multi-Robot Formation Navigation and Adaptation with Agentic LLMs

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER