NRTIRL Based NN-RRT* Path Planner in Human-Robot Interaction Environment

Abstract

In human-robot interaction environment, it is of great significance for mobile robots to have the awareness of social rules, to realize the socialization and anthropomorphism of robot navigation behavior, and enhance the scene adaptation ability of socialized navigation. Learning from Demonstration (LfD) can obtain an optimized robot trajectory by learning the expert path. Inspired by the LfD method, we propose a new navigation method to learn navigation behaviors form demonstration paths of experts by Neural Network Rapidly-exploring Random Trees (NN-RRT*) planner in the human-robot interaction environment. First, we propose a new NN-RRT* planner to generate paths. Next, the features of demonstration paths and planned paths are extracted for Inverse Reinforcement Learning (IRL) process. The cost function of the path planner is updated. Finally, a new NN-RRT* that can adapt to the complex human-robot interaction environment is obtained. The experimental results show that comparing with the state-of-the-art methods, the path generated by the new navigation method has a higher degree of anthropomorphism and is suitable for navigation in a complex human-robot interaction environment.