Space-Heuristic Navigation and Occupancy Map Prediction for Robot Autonomous Exploration

Ping Zhong, Bolei Chen, Yongzheng Cui, Hanchen Song, Yu Sheng

Abstract: Efficient autonomous exploration in unknown environments is a challenging and basic problem in the field of robotics. Due to the lack of priori information of the environment, the robot cannot clearly select the region with high information gain for exploration. In addition, the existing greedy exploration strategy leads to repeated back-and-forth movements of the robot and inefficient exploration. Intelligent bodies like humans often rely on past experience to infer the structural characteristics of the environment which are used to assist exploration. To give the robot spatial awareness, we propose a method of predicting map occupancy by using Generative Adversarial Networks (GANs). The adversarial loss and the feature extraction loss used in the training of GANs are introduced in detail. Further, we propose a novel utility function for the evaluation of exploration goals based on path length and structural characteristics of the environment. The utility function is used to guide the robot to efficiently explore step by step. We also propose a space-heuristic path planning method named CI-RRT* for robot navigation. We demonstrate the superiority of the proposed methods through comparison and ablation experiments in simulation environments. The experimental results prove that our method is superior to the existing methods.