A fast robot path planning algorithm based on bidirectional associative learning. (May 2021)

• Record Type: Journal Article
• Authors: Zhao, Meng; Lu, Hui; Yang, Siyi; Guo, Yinan; Guo, Fengjuan
• Is Part Of: Computers & industrial engineering. Volume 155(2021)
• Publication Date: 2021-05
• Subjects: Path planning -- Bidirectional associative learning -- Experience guidance -- Search scope -- Planning efficiency; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107173
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A robot fast path planning algorithm used in unknown environment is proposed. The episode is defined to satisfy the continuity of robot position in application. Reducing planning time by locking search scope in the early stage of planning. Some action selection strategies are proposed to improve the planning efficiency. Abstract: Fast path planning in unknown environment is important to reduce the loss of human and material resources. To reduce planning time while obtaining a short path, this paper proposes a Bidirectional Associative Learning Algorithm (BALA). In the proposed algorithm, an episode is defined as a bidirectional movement between the start point and the target point. The planning process in the BALA is divided into three stages: early stage, medium stage and end stage. In the early stage, the attraction of the target point is adopted to instruct the robot to select action. This strategy not only helps the robot avoid blind search, but also provides the search scope that may contain the global shortest path for the subsequent episodes. In the medium stage, we propose an action selection strategy based on the experience guidance, where the experience obtained in the obverse and reverse movements is used alternately to improve the learning efficiency of the robot. In the end stage, a strong connectivity relationship between nodes is defined. Planning by this relationship, the length of the final planned path will be the shortest based on the experience the robot obtains. The comparison results with Q-Learning and its improved algorithm reveal that the BALA demonstrates desirable and stable performance in planning efficiency in any environment, and it can well balance the planning time and path length. Additionally, the practicability of the proposed algorithm is validated on Turtlebot3 burger robot.