A Unified Receding-Horizon Controller for Safe Multi-Robot Search and Rescue, Pursuit, and Perimeter Defense
Abstract
Coordinating a robot team safely in real time means deciding on board within a hard cycle. We present a receding-horizon controller in which each non-holonomic robot solves its own short-horizon subproblem every cycle, under an idealized shared-state coordination layer. One cost function covers cooperative pursuit, perimeter defense and search and rescue (SAR). We ask when the per-cycle optimizer, and the computation given to it, change the mission outcome. All results are simulated, with static disk obstacles and, outside a dedicated test, noise-free sensing, ideal communication and instantaneous actuation. Six hypotheses were fixed in advance and tested on five random layouts per task and by interventions; three held. At full budget, five optimization-based controllers never touched an obstacle under a continuous-time check, whereas a potential-field planner did in 35% of random-layout episodes; unmodeled actuator lag did cause contacts. Swapping the optimizer among SAM-SOMA, PSO, DE, GWO, L-SHADE and a sampling MPC (MPPI) changed pursuit time by a few percent on random layouts. In SAR below 700 evaluations, SAM-SOMA and DE were faster than PSO and GWO on average. At matched end-to-end time no solver was better across tasks and layouts. On the fixed SAR layout, more computation past 1000 evaluations slowed the mission: the yaw-effort penalty charges a heading-locked sensor for the turning it needs, and removing it, or widening the sensor, removed the effect. On random layouts this slowdown did not appear on average. At 700 evaluations the mean five-robot cycle used at most 19% of the 120 ms period.