Data-Driven Reachability Analysis of Pedestrians Using Behavior Modes

TL;DR

This paper introduces a modular, data-driven method for pedestrian reachability analysis that leverages behavior modes to produce less conservative, safety-guaranteed predictions of future pedestrian states.

Contribution

It proposes a novel approach that splits pedestrian trajectories into behavior modes for improved, less conservative reachability analysis with safety guarantees.

Findings

Modal reachable sets are less conservative.

The approach is effective on real pedestrian data.

Supports different behavior splitting methods.

Abstract

In this paper, we present a data-driven approach for safely predicting the future state sets of pedestrians. Previous approaches to predicting the future state sets of pedestrians either do not provide safety guarantees or are overly conservative. Moreover, an additional challenge is the selection or identification of a model that sufficiently captures the motion of pedestrians. To address these issues, this paper introduces the idea of splitting previously collected, historical pedestrian trajectories into different behavior modes for performing data-driven reachability analysis. Through this proposed approach, we are able to use data-driven reachability analysis to capture the future state sets of pedestrians, while being less conservative and still maintaining safety guarantees. Furthermore, this approach is modular and can support different approaches for behavior splitting. To illustrate the efficacy of the approach, we implement our method with a basic behavior-splitting module and evaluate the implementation on an open-source data set of real pedestrian trajectories. In this evaluation, we find that the modal reachable sets are less conservative and more descriptive of the future state sets of the pedestrian.