Robotic Full-Coverage Path Planning and Local Path Tracking Control for Cleaning in Complex Environments

Abstract

To address the issues of fragmented paths, frequent turning, and missed coverage caused by conservative obstacle avoidance in complete-coverage path planning (CCPP) for cleaning robots in complex environments with static obstacles, this paper proposes a collaborative navigation framework that integrates global coverage planning and local tracking control into a unified architecture. At the global planning level, mechanisms including region decomposition and merging, optimal coverage direction selection and maintenance, an obstacle-crossing strategy, and an improved dead-zone escape mechanism are introduced based on the Bio-inspired Neural Network (BINN) to enhance coverage quality. Specifically, the region decomposition and merging, combined with the optimal direction selection mechanism, ensure regular coverage along the optimal direction within each sub-region. Furthermore, by relying on the local layer’s autonomous obstacle avoidance capability, the obstacle-crossing strategy intentionally ignores minor obstacles to prevent interference with neural activity caused by avoidance maneuvers, thereby avoiding subsequent tortuous paths and frequent turns, which further enhances overall operational efficiency. At the local planning level, a dual-layer controller architecture is developed. A local path controller fusing Model Predictive Control (MPC) and the Dynamic Window Approach (DWA) achieves high-precision path tracking and collision detection. When a collision risk is detected, a local path replanner re-optimizes the global path segments to enable minimal-amplitude, edge-following obstacle avoidance and rapid return to the global path, thereby preventing under-coverage and redundant coverage caused by excessive detouring. Simulation and real-world experiments conducted in a courtyard environment demonstrate that the redundant coverage rate in global planning is reduced to 7.01%, outperforming traditional methods in total path length and accumulated turning angles. For local planning, the Integral of Absolute Error (IAE) is 18.3296, which is 26.12% and 27.15% lower than those of the Timed Elastic Band (TEB) and DWA algorithms, respectively, while achieving a coverage rate of 91.16%. Furthermore, the overall discrepancy between real-world and simulation results is within 6%, validating the feasibility and effectiveness of the proposed method for practical cleaning tasks.