Robustness in trajectory prediction for autonomous vehicles: a survey

Autonomous vehicles rely on accurate trajectory prediction to inform decision-making processes related to navigation and collision avoidance. However, current trajectory prediction models show signs of overfitting, which may lead to unsafe or suboptimal behavior. To address these challenges, this paper presents a comprehensive framework that categorizes and assesses the definitions and strategies used in the literature on evaluating and improving the robustness of trajectory prediction models. This involves a detailed exploration of various approaches, including data slicing methods, perturbation techniques, model architecture changes, and post-training adjustments. In the literature, we see many promising methods for increasing robustness, which are necessary for safe and reliable autonomous driving.

I. Introduction

A critical technological aspect in the control of autonomous vehicles is forecasting future states of the surrounding environment to facilitate safe navigation through dynamic surroundings. The predicted future trajectories of traffic participants play an important role in detecting potential hazards in advance and are essential for the design of decision-making or planning algorithms [1], [2]. This predictive task relies on utilizing perceived data, such as camera and LiDAR data [3]-[6]. This data is mostly tracked across multiple timestamps, transformed into trajectories, and then integrated into maps to enable accurate predictions [3], [5].