“To Approach Humans?”: A Unified Framework for Approaching Pose Prediction and Socially Aware Robot Navigation

Abstract

We propose a unified framework for approaching pose prediction, and socially aware robot navigation, which enables a mobile service robot to safely and socially approach a dynamic human or human group in a social environment. The proposed framework is composed of four major functional blocks: 1) human detection and human features extraction to estimate the human states, and the social interaction information from the socio-spatio-temporal characteristics of a human and a group of humans; 2) a dynamic social zone (DSZ) consisting of an extended personal space and a social interaction space is modeled by the human states and social interaction information to represent space around the human and human group; 3) the approaching pose of the robot to a human or a human group is predicted using the DSZ and the environmental surroundings; and 4) the DSZ and the estimated approaching pose are incorporated into a motion planning system, comprising a local path planner and dynamic window approach technique, to generate the motion control commands for the mobile robot. We evaluate the developed framework through both simulation and real-world experiments under the newly proposed human safety and comfort indices, including the social individual index, social group index, and social direction index. The results show that the unified framework is fully capable of driving a mobile robot to approach both stationary and moving humans and human groups in a socially acceptable manner while guaranteeing human safety and comfort.