Real-Time Multiple Obstacle Avoidance and Navigation in Unstructured Spaces with Applications to Indoor Delivery Robots
Yuhui Wang, Unnikrishna S. Pillai, Juntao Chen, Junaid Farooq · 2025
The growing deployment of autonomous delivery robots in indoor environments, such as offices and corporate buildings, demands robust navigation systems capable of realtime decision-making in dynamic and unstructured spaces. These robots must operate efficiently without prior knowledge of the environment, while avoiding static and dynamic obstacles, including walls, furniture, and human traffic. To address these challenges, this paper presents a real-time navigation framework based on a modified artificial potential field (APF) method tailored for unstructured indoor spaces. The proposed approach integrates two key enhancements: object proximity-seeking, inspired by animal motion behavior, and trace memory, which dynamically mitigates the local minimum problem and prevents redundant path exploration. The object proximity-seeking mechanism enables the robot to maintain safe distances from nearby obstacles, while the trace memory generates repulsive forces from previously visited locations, encouraging efficient space exploration and reducing zigzag trajectories. The proposed framework requires no global map or GPS support, relying solely on onboard depth camera sensing. Through extensive simulations and real-world experiments conducted in an office environment, we demonstrate that the robot effectively generates smooth, collision-free trajectories towards the destination and significantly improve navigation efficiency in cluttered and dynamic indoor spaces.