SafeShift: Safety-Informed Distribution Shifts for Robust Trajectory Prediction in Autonomous Driving

TL;DR

This paper introduces SafeShift, a framework for identifying safety-critical scenarios within real-world datasets to evaluate and improve the robustness of trajectory prediction models in autonomous driving.

Contribution

It presents a novel safety-informed distribution shift framework, including scenario characterization, a scoring scheme for risky scenarios, and a remediation strategy to enhance model safety.

Findings

10% reduction in prediction collision rates

Effective identification of safety-critical scenarios

Improved robustness of trajectory prediction models

Abstract

As autonomous driving technology matures, safety and robustness of its key components, including trajectory prediction, is vital. Though real-world datasets, such as Waymo Open Motion, provide realistic recorded scenarios for model development, they often lack truly safety-critical situations. Rather than utilizing unrealistic simulation or dangerous real-world testing, we instead propose a framework to characterize such datasets and find hidden safety-relevant scenarios within. Our approach expands the spectrum of safety-relevance, allowing us to study trajectory prediction models under a safety-informed, distribution shift setting. We contribute a generalized scenario characterization method, a novel scoring scheme to find subtly-avoided risky scenarios, and an evaluation of trajectory prediction models in this setting. We further contribute a remediation strategy, achieving a 10% average reduction in prediction collision rates. To facilitate future research, we release our code to the public: github.com/cmubig/SafeShift