A Hybrid Multi-Goal Path Planning Framework for
Powered Wheelchairs in Dynamic Environments:
From Simulation to Real-World Validation

Abstract

Autonomous powered wheelchairs (PWs) operating in human-centered environments must combine efficient multi-destination path planning with continuous collision avoidance strategies in the presence of static and dynamic obstacles. This paper presents a hybrid framework integrating Traveling Salesman Problem (TSP)-based global task ordering, Potential-Field (PF) guidance, and local Rapidly-Exploring Random Tree (RRT) and A* planners. PF guidance transfers global path information to the local planning layer and helps the robot remain aligned with the global path during obstacle avoidance. The simulation model uses an elliptical wheelchair representation to account for the platform dimensions during collision avoidance. The local planning layer is enhanced through an extended-neighborhood, direction-aware A* formulation or an RRT planner with weighted environment-aware sampling and a two-step extension strategy. The TSP–PF–RRT and TSP–PF–A* methods are evaluated in normal, crowded, and extra-crowded environments using both point-based and elliptical robot models. The framework is implemented on a physical PW and evaluated using motion-capture (MoCap) for ground-truth measurements and onboard LiDAR and RGB-D sensing without external MoCap support. A representative real-world accessibility scenario is additionally investigated in simulation and hardware. Results show that incorporating the wheelchair geometry changes the observed collision behavior, yielding comparable collision avoidance performance between the two hybrid planners under the elliptical model.