Improved DWA Algorithm for Mobile Robot Obstacle Avoidance through Multi-Sensor Information Fusion

Aiming at the dynamic feasibility of mobile robots passing through complex moving obstacles under global navigation and their current spatial perception capabilities, as well as the prediction of target behavior paths, a novel improved DWA obstacle avoidance strategy based on the fusion of visual information and radar parameters, termed DWA-MSF, is proposed. This method combines the advantages of the Dynamic Window Approach (DWA) while satisfying the known kinematic constraints of the robot. The visual sensor utilizes the YOLOv8 object detection and classification algorithm, combined with laser beam information from the LiDAR to collect obstacle point cloud information. This allows for rapid processing of unknown moving obstacles without affecting the smoothness of the robot's path. Furthermore, it predicts the basic movement direction and orientation changes of obstacles in advance. The sensor-collected information is processed into numerical parameters, which are then introduced into the DWA algorithm as collision risk sub-functions, thereby avoiding using the detected target information points as local moving target points for the DWA algorithm. This dynamically adjusts the robot's motion trajectory to reduce collision risks. The DWA algorithm searches for the optimal action commands in the robot's velocity and angular velocity space, ensuring the robot avoids detected obstacle trajectories. Through a series of simulation experiments based on MATLAB and ROS, the results demonstrate that in complex obstacle environments, the proposed DWA-MSF algorithm reduces the change in heading angle by 30.7% and the path length by 21.3% when predicting obstacle types and trajectory paths in advance.