Heterogeneous Decision Making in Mixed Traffic: Uncertainty-aware Planning and Bounded Rationality

TL;DR

The paper studies heterogeneous decision making in mixed traffic where human drivers exhibit bounded rationality and automated vehicles (AVs) perform uncertainty-aware planning. It introduces a two-agent framework with HV short-horizon planning and AV $L$-step lookahead planning based on predictions of HV actions, incorporating a Gaussian prediction error to capture uncertainty. The authors derive regret bounds for both AV and HV in linear and nonlinear dynamics, revealing error accumulation and Goodhart-like effects as the planning horizon grows and prediction accuracy varies. They further aggregate these results into a system-level regret bound and provide empirical insights, offering guidance on choosing planning horizons and improving prediction models to enhance mixed-autonomy safety and efficiency.

Abstract

The past few years have witnessed a rapid growth of the deployment of automated vehicles (AVs). Clearly, AVs and human-driven vehicles (HVs) will co-exist for many years, and AVs will have to operate around HVs, pedestrians, cyclists, and more, calling for fundamental breakthroughs in AI designed for mixed traffic to achieve mixed autonomy. Thus motivated, we study heterogeneous decision making by AVs and HVs in a mixed traffic environment, aiming to capture the interactions between human and machine decision-making and develop an AI foundation that enables vehicles to operate safely and efficiently. There are a number of challenges to achieve mixed autonomy, including 1) humans drivers make driving decisions with bounded rationality, and it remains open to develop accurate models for HVs' decision making; and 2) uncertainty-aware planning plays a critical role for AVs to take safety maneuvers in response to the human behavior. In this paper, we introduce a formulation of AV-HV interaction, where the HV makes decisions with bounded rationality and the AV employs uncertainty-aware planning based on the prediction on HV's future actions. We conduct a comprehensive analysis on AV and HV's learning regret to answer the questions: 1) {How does the learning performance depend on HV's bounded rationality and AV's planning}; 2) {How do different decision making strategies impact the overall learning performance}? Our findings reveal some intriguing phenomena, such as Goodhart's Law in AV's learning performance and compounding effects in HV's decision making process. By examining the dynamics of the regrets, we gain insights into the interplay between human and machine decision making.

Figures (4)

• Figure 1: Numerical results on AV's regret. (a) The impact of planning horizon $L$ on AV's performance gap (ref. Lemma \ref{['lemma:av2']}). (b) The impact of the planning horizon $L$ on AV's regret $\mathcal{R}_A$. (c) The impact of planning horizon on regret dynamics $\mathcal{R}_A(T)$ during the interactions.
• Figure 2: Empirical studies on AV and HV's decision making on the overall performance.
• Figure 3: Empirical results on how AV and HV's decision making have impact on the overall regret dynamics, i.e., take regret as function of $T$.
• Figure 4: Illustration of the impact of $\mu_A$, $\mu_H$ on (a) the regret summation and (b) the overall regret.

Theorems & Definitions (8)

• Remark 3.1
• Remark 4.1: Generalization of the Prediction Error Assumption
• Lemma 4.3: AV's Performance Gap in the Linear Case.
• Lemma 4.4: AV's Performance Gap in Non-linear Case
• Theorem 4.6: Regret on AV's Decision Making
• Theorem 4.7: Regret for HV.
• Corollary 5.1: Regret of the HV-AV Interaction System
• Remark 5.2: Extension beyond two-agent case