A probabilistic model of human motion and navigation intent for mobile robot path planning

TLDR

Robots require accurate human motion prediction for path planning, yet common methods assume constant velocity, which is often invalid. This study aims to infer human navigation intent and use it to predict motion. The authors construct a probabilistic motion model that combines manually defined functional places, automatically extracted way‑points, and motion‑probability grids, and integrate it with a laser‑sensing robot platform, validating it on real office‑environment data. The model successfully predicts human trajectories, as illustrated by example predictions.

Abstract

In order to effectively plan paths in environments inhabited by humans, robots must accurately predict human motion. Typical approaches to human prediction simply assume a constant velocity which is not always valid. This paper proposes to determine the likely navigation intent of humans and use that to predict human motion. Navigation intent is determined by the function and structure of the environment. Manually assigned functional places are combined with automatically extracted navigation way-points to define a number of likely navigation targets within the environment. To predict human motion toward these targets, a probabilistic model of human motion is proposed which is based on motion probability grids generated from observed motion. The models of human navigation intent and motion are integrated with an autonomous mobile robot system, with a laser range sensor detecting humans moving within the environment, and a path planning system. The models of human navigation intent and motion are verified using real captured human motion data from an office environment. Examples of human motion prediction are also presented.