Vehicle trajectory prediction works, but not everywhere

TL;DR

This paper demonstrates that current vehicle trajectory prediction models do not generalize well to new scenes, and introduces an adversarial scene generation method that exposes their vulnerabilities and improves robustness.

Contribution

The paper presents a novel adversarial scene generation approach that reveals the limitations of existing models and enhances their robustness against off-road prediction failures.

Findings

Over 60% of benchmark scenes can be modified to cause off-road failures.

Generated scenes are realistic and reflect real-world scenarios.

Robustness of models improves by 30-40% after training with adversarial scenes.

Abstract

Vehicle trajectory prediction is nowadays a fundamental pillar of self-driving cars. Both the industry and research communities have acknowledged the need for such a pillar by providing public benchmarks. While state-of-the-art methods are impressive, i.e., they have no off-road prediction, their generalization to cities outside of the benchmark remains unexplored. In this work, we show that those methods do not generalize to new scenes. We present a method that automatically generates realistic scenes causing state-of-the-art models to go off-road. We frame the problem through the lens of adversarial scene generation. The method is a simple yet effective generative model based on atomic scene generation functions along with physical constraints. Our experiments show that more than 60% of existing scenes from the current benchmarks can be modified in a way to make prediction methods fail (i.e., predicting off-road). We further show that the generated scenes (i) are realistic since they do exist in the real world, and (ii) can be used to make existing models more robust, yielding 30-40 reductions in the off-road rate. The code is available online: https://s-attack.github.io/.