Knowledge-driven path planning for mobile robots: relative state tree

Path planning is important in the field of mobile robot. However, traditional path planning techniques optimize the navigation path solely based on the models of the robot and the environments. Owing to the time-varying environment, the robot is expected to launch the replanning procedure in real-time continuously. It is slow and wastes computing resources for repeated decisions. In this study, a new perspective is adopted which utilizes a knowledge-driven approach for path planning. The concept of relative state tree is proposed to develop an incremental learning method based on a path planning knowledge base. The knowledge library, which stores a collection of the mappings from environmental information to robot decisions, can be established by offline or online learnings. As the robot plans online, its movement is guided by the optimal decision that is retrieved from the library based on the information which matches mostly the current environment. A large number of simulations are executed to verify the proposed method. When comparing to \(k\)-d tree, this novel method has shown to use smaller storage space and have higher efficiency.