ANMIP: Adaptive Navigation based on Mutual Information Perception in Uncertain Environments.

Navigation in uncertain environment has become a hot research topic. Some path planning algorithms have been proposed to address the uncertainty problem, such as Canadian Traveller’s Problem (CTP) algorithm. However, these algorithms usually require accurate environmental information, which is often difficult to obtain. In addition, their decisions are based on robot current perceptions and the previous navigation experience is usually ignored, whereas the past experience is an important reference for people to optimize policies. In order to address these issues, we propose a mutual information perception based navigation method for efficient navigation in uncertain environment. First, an efficient CTP solver is proposed to quickly generate policy based on the obtained environment information. Second, a topological map representation method is proposed for map decomposition. In order to perceive environmental information, a block judgment interface module is proposed. Then, the door edge resolver algorithm is proposed to absorb the experience of the previous navigation. Finally, we design a complete information updating mechanism based on Wilson confidence interval, so that the robot can update its perception of the environment and realize adaptive navigation in uncertain environments. The experimental results show that by comparing with the existing move_base navigation system, our method has better performance in average navigation cost and navigation success rate.