Autonomous Obstacle Avoidance Motion Planning for Mobile Robots in Dark and Weak Environments Based on Multimodal Information Fusion

Mobile robots face the problems of weak visual perception and difficult navigation when operating in dark and low-visibility environments, which makes it challenging to avoid obstacles effectively. In this project, a multimodal information fusion-based autonomous motion planning method for obstacle avoidance is designed to enable autonomous obstacle avoidance in dark and low-visibility environments for mobile robots. First, the mobile robot selects a vision camera, LIDAR, and IMU sensors to measure multimodal information, thereby enhancing its perception of different environmental conditions. Then, the dual-stream network architecture is applied to ensure the effective encoding and decoding of multi-scale features to realize the multimodal information fusion based on the cross-attention mechanism. Based on the multimodal fusion information, an improved ant colony algorithm embedded in a dynamics model is used to design the robot trajectory. Finally, the theory proposed in this paper is validated using virtual and experimental prototypes. The results show that the method has an accuracy of more than 90% for obstacle targets in dark and weak environments, and enables the robot to avoid obstacles smoothly during motion.