Portal de Periódicos da CAPES

Motion Saliency-Based Collision Avoidance for Mobile Robots in Dynamic Environments

2021; Institute of Electrical and Electronics Engineers; Volume: 69; Issue: 12 Linguagem: Inglês

10.1109/tie.2021.3128885

1557-9948

Binghua Guo, Nan Guo, Zhisong Cen,

Multimodal Machine Learning Applications

Obstacle avoidance is a sizable challenge for robots working in a multiple dynamic obstacle environment. Conventional obstacle avoidance methods often require complex calculations to process all dynamic obstacles detected in the scene. Avoiding dangerous moving objects in time is often difficult when obstacles are many. In this article, we propose a robot obstacle avoidance method based on motion saliency for dynamic environments. First, we use segmented dynamic objects to calculate the saliency of dynamic objects and segment dangerous dynamic objects. Then, we use the B-spline curve to predict the movement of dangerous dynamic objects and combine it with the nonlinear model predictive control method to avoid dangerous obstacles in the dynamic environment of the robot. Considering the motion behavior of different dynamic objects, our obstacle avoidance strategy is to generate obstacle-free paths by adjusting the speed of the robot or connecting the centers of the rolling variable-size circles. Finally, we conduct a series of experiments to verify the effectiveness of our method.