Simultaneous people tracking and robot localization in dynamic social spaces

Abstract

Accurate robot localization and people tracking are necessary for deploying service robots in crowded everyday environments such as shopping malls, and features like product displays change over time, making map-based localization using on-board sensors difficult. We propose the use of an external sensor system to track people together with one or more robots. This approach is more robust to occlusions than on-board sensing and is unaffected by changing map features. In our system, laser range finders track people and robots in the environment, and odometry data is used to associate each robot with a tracked entity and correct the robot's pose. Techniques are also presented for identifying and recovering from tracking errors. Simulation results show that our system can outperform localization using on-board sensors, both in tracking accuracy and in automatic recovery from errors. We demonstrate our system's effectiveness in simulation, in a controlled experiment in a real shopping mall environment, and in real human---robot interactions with customers in a busy shopping arcade.