FastFunnels: A New Method for High-Speed Navigation of Multiple Robots in Narrow Spaces

Vignesh Rajagopal and Rohan Chandra propose FastFunnels, a hierarchical multi-robot navigation method: at the high level, a reinforcement learning policy generates deformable elliptical envelopes in real time, and at the low level, distributed model predictive control keeps robots within the envelopes and avoids collisions with one another. The goal is to enable Ackermann-steered multi-robot systems to travel at high speed through narrow spaces. In 14 simulated F1TENTH racing scenarios (1 to 6 robots, with the average of 20 trials per scenario), the authors report that FastFunnels achieves an 89% completion rate at 7.73 m/s, whereas all baselines either fail to complete the track or complete it only at low speeds of 0.07–1.79 m/s, with a completion rate of about 40%. These results come from simulation evaluation and have not yet been validated on real robots.

AI 根据报道生成 · 2 days ago更新