Dynamic Mapping and 3D Perception Using Voxel Grid and Modified Artificial Potential Fields for Indoor Locomotion

Abstract

This paper proposes an advanced 3D indoor navigation system for a mobile robot. The proposed method integrates RTAB-Map with Voxel Grid Filters and Joint Probabilistic Data Association (JPDA) to generate surrounding environment map efficiency. Additionally, the local path planner combines pure pursuit with a modified Artificial Potential Field (APF) method to improve navigation capability. It generates steering commands and desired velocities and adjusts the attractive potential force equation to maintain balance and operational efficiency. This modification improves safety, pedestrian avoidance, and comfort by minimizing unnecessary rotations while ensuring smooth navigation. The proposed system improves the locomotion ability by reducing roll, pitch, and yaw fluctuations by approximately 30% compared to traditional APF methods. Voxel grid filtering enhances computational efficiency, reducing processing time per iteration by up to 73% - from 0.247 seconds (raw LiDAR) to 0.067 seconds (voxel size of 0.9) - while maintaining obstacle detection accuracy. The integration of JPDA ensures safe multi-target detection, with minimum safe distances of 0.94 meters from dynamic actors and a Threat Level Index (TLI) peaking at 0.24. In a scenario comparing two robots with different map knowledge, the robot with map knowledge reached the waypoint 20% faster, following an efficient path. However, despite lacking prior knowledge, the second robot reached the waypoint, demonstrating the system’s adaptability. These quantitative results confirm the proposed method’s capability to enhance safety, efficiency, and human comfort, making it suitable for real-time indoor navigation in dynamic environments.