Robot navigation in dense human crowds: Statistical models and experimental studies of human–robot cooperation

Abstract

We consider the problem of navigating a mobile robot through dense human crowds. We begin by exploring a fundamental impediment to classical motion planning algorithms called the ''freezing robot problem'': once the environment surpasses a certain level of dynamic complexity, the planner decides that all forward paths are unsafe, and the robot freezes in place (or performs unnecessary maneuvers) to avoid collisions. We argue that this problem can be avoided if the robot anticipates human cooperation, and accordingly we develop interacting Gaussian processes, a prediction density that captures cooperative collision avoidance, and a ''multiple goal'' extension that models the goal-driven nature of human decision making. We validate this model with an empirical study of robot navigation in dense human crowds (488 runs), specifically testing how cooperation models effect navigation performance. The multiple goal interacting Gaussian processes algorithm performs comparably with human teleoperators in crowd densities nearing 0.8 humans/m 2 , while a state-of-theart non-cooperative planner exhibits unsafe behavior more than three times as often as the multiple goal extension, and twice as often as the basic interacting Gaussian process approach. Furthermore, a reactive planner based on the widely used dynamic window approach proves insufficient for crowd densities above 0.55 people/m 2 . We also show that our noncooperative planner or our reactive planner capture the salient characteristics of nearly any dynamic navigation algorithm. Based on these experimental results and theoretical observations, we conclude that a cooperation model is critical for safe and efficient robot navigation in dense human crowds.

Planning and Decision-Making for Autonomous Vehicles

“…However, even under perfect individual prediction and perfect knowledge of all agents' trajectories, the freezing-robot problem cannot always be prevented (85).…”

“…In interacting Gaussian processes (85), each agent's trajectory is modeled via a Gaussian process. Individual Gaussian processes are coupled through an interaction potential that models cooperation between different agents' trajectories.…”

Group abstraction for assisted navigation of social activities in intelligent environments

“…Under these circumstances, any actual deviations to the plan tend to be minor and are treated as sensing errors or slippage, etc. This can actually cause a robot to "freeze", especially in chaotic and busy environments, such as those for which the ACANTO device is intended [38].…”

“…Somewhat similar work is considered in [38], where the goal is for a robot to act collaboratively with pedestrians in a crowded environment. However, the approach requires significant prior knowledge about the (very small) environment and pedestrians.…”

Breaking the Human-Robot Deadlock: Surpassing Shared Control Performance Limits with Sparse Human-Robot Interaction

“…We thus introduce generalized shared control (GSC), where instead of arbitrating over predictions from the human and robot models, we arbitrate over the full human and robot distributions themselves (bottom half of Figure 1). Our approach is motivated by the following: distributions represented in a Gaussian process (GP) basis [49], [54], [55], [5], [30] explicitly capture multifaceted intentionality (which we call "intention ambiguity") as well as an agent's willingness to compromise within a given strategy (which we call "flexibility"). In Figure 2, we illustrate intention ambiguity and flexibility, and in Section III we define intention ambiguity and flexibility.…”