Abstract:
Constructing realistic, real time, human-robot interaction models is a core challenge in crowd navigation. In this paper we derive an interaction density (based on multi-modal Gaussian process agent models) from first principles of probability theory; we call our approach “optimization shaped interacting Gaussian processes” (osIGP). Furthermore, we compute locally optimal solutions—with respect to multi-faceted agent “intent” and “flexibility”—in near real time on a laptop CPU. We test on challenging scenarios from the ETH crowd dataset and show that the safety and efficiency statistics of osIGP is indistinguishable from human safety and efficiency statistics. Further, we compute the safety and efficiency statistics of dynamic window avoidance, a physics based model variant of osIGP, and the best performing deep reinforcement learning approach; osIGP outperforms all of them. Finally, we show substantial improvement over a Monte Carlo based approach.
