Detecting Static Conversation Group with Common Concern Area for Robot Navigation and Behavior Evaluation

An F-formation detection algorithm based on common concern areas is proposed for static conversation group. It takes the position and direction of pedestrians as input to construct the common concern area of groups. Then a sliding window-based maximum filter is used to detect the group center for clustering. After detecting the static conversation group, a group comfort space is constructed for a time-dependent A^* path planning algorithm based on multi-layer costmap mechanism. Thus, mobile robot navigation considering group comfort is realized. In addition, it is challenging to quantitatively evaluate whether a robot's navigation behavior is socially acceptable. A graph convolutional network-based model for evaluating robot behavior is constructed. Experimental results show that the evaluation network has a similar capacity to humans in evaluating the robot behavior. Visualization results show the rationality of the evaluation network's results. According to the evaluation network, robots considering static conversation group can produce more comfortable trajectories.