Autonomous navigation experiment for mobile robot based on IHDR algorithm

This paper investigates the robot navigation based on visual servo and learning algorithm. A novel method for image compression and noise reduction is proposed for constructing the IHDR (incremental hierarchical discriminant regression) tree under the real indoor environment. An autonomous learning framework which consists of three parts is constructed in the robot by using the method, i.e. knowledge learning, knowledge retrieval, and online updating. Firstly, we need to drive the robot to move around along with the same path for several times, meanwhile, recording the videos using a fixed camera on the robot. The robot velocity and the videos are recorded at the same frequency. Secondly, by using the IHDR algorithm, the IHDR tree is established through the collected samples, and it is a mapping relationship between input (images information) and output (robot velocity). Finally, the robot could decide to select the corresponding velocity from the knowledge library according to the most matched environment for each online scene to realize robot autonomous navigation. The experiment and simulation results illustrate that the proposed method can be well applied in the robot navigation, and the more repetition of training samples, the better performance of autonomous navigation. Furthermore, this method also has capacity of resisting disturbance even with the Gaussian noise in the images.