Occlusion-Aware Crowd Navigation Using People as Sensors

TL;DR

This paper introduces an occlusion-aware navigation method for robots in crowded environments, leveraging social inference and deep learning to estimate hidden obstacles, achieving effective real-world navigation transfer.

Contribution

It presents the first use of social occlusion inference in crowd navigation, integrating variational autoencoders and deep reinforcement learning for improved obstacle estimation.

Findings

Achieves comparable collision avoidance to fully observable methods in simulation.

Successfully transfers occlusion-aware policy from simulation to real-world Turtlebot 2i.

Demonstrates the effectiveness of social inference techniques in dynamic, occluded environments.

Abstract

Autonomous navigation in crowded spaces poses a challenge for mobile robots due to the highly dynamic, partially observable environment. Occlusions are highly prevalent in such settings due to a limited sensor field of view and obstructing human agents. Previous work has shown that observed interactive behaviors of human agents can be used to estimate potential obstacles despite occlusions. We propose integrating such social inference techniques into the planning pipeline. We use a variational autoencoder with a specially designed loss function to learn representations that are meaningful for occlusion inference. This work adopts a deep reinforcement learning approach to incorporate the learned representation for occlusion-aware planning. In simulation, our occlusion-aware policy achieves comparable collision avoidance performance to fully observable navigation by estimating agents in occluded spaces. We demonstrate successful policy transfer from simulation to the real-world Turtlebot 2i. To the best of our knowledge, this work is the first to use social occlusion inference for crowd navigation.