Functional Architecture and Implementation of an Autonomous Emergency Steering System

TL;DR

This paper presents a comprehensive software architecture for autonomous emergency steering systems that estimates vehicle capabilities, plans and assesses paths, and triggers maneuvers based on a novel time-to-evade metric, validated through simulations and field tests.

Contribution

It introduces a complete, tractable architecture for AES with a novel TTE-based triggering methodology and adaptive replanning under uncertainties.

Findings

The architecture effectively reduces collision risk in simulations.

Field tests confirm the system's real-world applicability.

Replanning improves safety under measurement uncertainties.

Abstract

Autonomous emergency steering (AES) systems have the promising potential to further reduce traffic fatalities with other (potentially vulnerable) traffic participants by using relatively small lateral deviations to realize collision-free behavior. In this work, a complete and tractable software architecture is presented for such an AES system, comprising of the estimation of the vehicles capabilities, planning a set of paths which exploit these capabilities, checking the feasibility and risk of these paths and eventually triggering the decision to drive along one of these paths, i.e., initiating an AES manoeuvre. A novel methodology is provided to trigger such an AES system, which is based on a time-to-evade (TTE) notion. In the presence of time-varying uncertainties or measurement inaccuracies, the system is able to replan the path from the previously chosen path to ensure collision-free behavior. The proposed architecture and control approach is validated using a simulation study and field tests, showing the effectiveness of the architecture and its sub-components.