Fast Autonomous Exploration in Complex Environments via the Farthest Cluster Representative and Dynamic Information Gain

Junhui Kuang, Dezhi Zheng, Youyi Huang, Jianye Fu, Shubin Cai, Yinghui Pan, Zhong Ming

Abstract: Complex and dynamic exploration environments pose significant challenges to single-robot autonomous exploration algorithms. Existing methods often fail to address the issue of local area legacy, resulting in the robot’s backtracking. To tackle this problem, we introduce a novel potential field-based autonomous exploration framework that leverages Farthest Cluster Representative Selection (FCRS) and incorporates a Region-Aware Dynamic Information Gain (RADIG) approach to enhance exploration efficiency. We select the point farthest from the robot in each cluster as a candidate target point via the FCRS. In the RADIG module, we dynamically adjust the information gain of the unexplored areas. Specifically, we train a neural network dedicated to identifying leftover regions and increasing their information gain to enhance exploration priority. The combination of the FCRS and RADIG modules maximizes the range of each exploration and pays attention to corner areas. Extensive experiments on publicly available datasets demonstrate that the proposed framework significantly outperforms existing methods, reducing exploration time by 4.1% to 25.1% and path length by 1.8% to 28.3%, across various scenarios. The results indicate that these reductions become more pronounced as the complexity of the environment increases, highlighting the effectiveness of our approach.