Goal-directed pedestrian model for long-term motion prediction with application to robot motion planning

Abstract

A probabilistic goal-directed model is proposed for pedestrian motion using navigation function and statistics of human motion gathered in the environment. In comparison with existing models, this model is both computationally inexpensive and does not fail when the optimal direction of motion in terms of this model is non-unique. We further introduce a Rapidly-Exploring Random Tree (RRT)-based path planner developed for planning in state-time space. With the help of an improved distance metric, the planner is much faster than RRT-Blossom [11] in complex maps. In an environment with 10 pedestrians, the planner and motion prediction combined can perform in near-real time.