Development of a Memory-Based Anti-Oscillation Strategy to Improve Potential Field Path Planning for Robot Swarms

Abstract

Potential-field navigation in leader–follower robot swarms often suffers from oscillation and local trapping near obstacles, reducing goal-reaching reliability. This paper proposes PF+GBM, a Potential Field (PF) planner augmented with a goal-direction bias and a short-term memory of recently visited cells to discourage back-and-forth motion during gradient-descent stepping. Unlike baseline PF, the proposed method adds this memory-based anti-oscillation rule while keeping computation lightweight. Experiments in a Webots simulation integrated with Robot Operating System 2 (ROS2) were conducted in corridor, single-obstacle, and random-obstacle scenarios (five runs each) with a fixed start and varied targets. PF+GBM reached the goal in 13/15 runs versus 6/15 for baseline PF, and reduced mean path length in the corridor and random-obstacle cases (about 6.6% and 16.9%), while maintaining comparable travel time in successful runs.