A machine learning-based mobile robot visual homing approach

Visual homing enables mobile robots to move towards a previously visited location solely based on panoramic vision sensors. In this paper, a SIFT-based visual homing approach incorporating machine learning is presented. The proposed approach can reduce the impact of inaccurate landmarks on the performance, and generate more precise home direction with simple model. The effectiveness of the proposed approach is verified on both panoramic image databases and actual mobile robot, experimental results reveal that compared to some traditional visual homing methods, the proposed approach exhibits better homing performance and adaptability in both static and dynamic environments.