Interactive Prediction and Decision-Making for Autonomous Vehicles: Online Active Learning With Traffic Entropy Minimization

Interacting with the surrounding road users is crucial for autonomous vehicles (AV). However, the inherent multimodality and uncertainties associated with traffic participants (TP) pose challenges in AVs’ prediction and decision-making (PnD). A primary challenge is adapting predictors trained on static offline datasets to the dynamic, diverse data streams encountered in reality. Secondly, utilizing one single forecast trajectory with the highest probability for decision-making contains potential risks as it neglects that even a small probability represents a subset of TP behaviors. Based on the existing prediction backbone, we propose an online learning approach incorporating pseudo-labels inferred from partial feedback as compensation for conventional methodologies, considering both the commonsense and personalization facets of driving. Drawing inspiration from the second law of thermodynamics, we propose to minimize microscopic traffic entropy as an additional objective in decision-making. This objective aims to reduce the chaos of traffic scenes, thus achieving more predictable future interactions and, conversely, making future decisions easier. Through real-time human-in-the-loop experiments, we quantifiably and comparably reveal that adopting one single trajectory without online learning in PnD is risky. However, this reliability is verified to be significantly improved by our proposed techniques, and the efficacy is further analyzed in a subsequent qualitative study. A static experiment transferring the prediction algorithm trained exclusively on Argoverse 2 to datasets including NGSIM, HighD, RounD, and NuScenes is also conducted, demonstrating that the proposed correction can effectively mitigate the gap between the datasets and real-world scenarios.

I. Introduction

Prediction and decision-making are critical for automated driving. Automated vehicles (AV) must continue to make decisions under dynamic uncertainties induced by the complex, intertwined, and unknown environment [1]. By emulating human driver functionalities, an AV algorithm contains modules to predict the movements of surrounding traffic participants (TPs) and make decisions afterward [2]. However, this task is challenging, even for human drivers, due to the uncertainties associated with TPs. The potential future movements of human TPs are inherently multimodal [3]; that is, all forecast future trajectories within physical limits are reasonable. This characteristic adds complexity to decision-making for both humans and machines in such uncertain environments. While recent advancements in deep learning have significantly enhanced fields like language models and computer vision, these improvements also boost the development of self-driving algorithms, particularly in higher-level modules such as perception, prediction, and decision-making. Nonetheless, many implementation issues remain unresolved in this dynamic, uncertain, and highly personalized context. We aim to address high-level uncertainties via personalized prediction and dynamic online learning. Additionally, unlike conventional methods that select the prediction with the highest probability for PnD, our approach incorporates a novel traffic entropy reward in decision-making, easing both future predictions and decisions.