Performance of a low-cost, human-inspired perception approach for dense moving crowd navigation

We describe a method for low-cost awareness of characteristics of dense, moving crowds such as group formation, personal space approximation, and occlusion compensation for use in navigating through crowds. It incorporates social expectations and is inspired by human perceptual processes. The approach uses a single Kinect to cluster all moving objects into groups, applies a 2D polygon projection in obscured regions, and a group personal space modeled using asymmetric Gaussians in order to inhibit certain socially inappropriate robot paths. This approach trades off detection of individual people for higher coverage and lower cost, while preserving high speed processing. A real-world evaluation of this approach showed good performance in comparison to an existing people detection approach. The projected polygon step captures significantly more people in the scene (77% vs. 80%) and supports group clustering in dense, complex scenarios. Examples are provided for group splitting and merging, dense crowds with obstructions, and cases where other approaches typically encounter difficulty.

I. Introduction

As mobile robots become more common in society their novelty and entertainment value will reduce, thus creating new challenges for robot navigation. Currently, independent robot travel produces unnatural crowd dynamics. Bystanders unfamiliar with robots are usually careful, suspicious, and concerned for their safety, thereby creating large spaces for robots to move through. Similarly, inappropriate or rude behavior by the robot is tolerated due to novelty. Asocial methods like linear “robotic” travel are usually effective but create a nuisance, leading to resentment and sometimes undesirable contact and malicious actions by nearby humans (e.g., [1]). Therefore, robots will soon need to travel in crowds in more socially appropriate methods.