Bilevel Multi-Armed Bandit-Based Hierarchical Reinforcement Learning for Interaction-Aware Self-Driving at Unsignalized Intersections

TL;DR

This paper introduces BiM-ACPPO, a hierarchical reinforcement learning framework using bilevel multi-armed bandits for interaction-aware decision-making at unsignalized intersections, improving safety and efficiency.

Contribution

It proposes a novel bilevel RL approach with dynamic curriculum learning and interaction-aware planning for autonomous driving at intersections.

Findings

Outperforms baseline methods in CARLA simulations

Demonstrates strong generalization in new urban scenarios

Enhances sample efficiency through dynamic curriculum adjustment

Abstract

In this work, we present BiM-ACPPO, a bilevel multi-armed bandit-based hierarchical reinforcement learning framework for interaction-aware decision-making and planning at unsignalized intersections. Essentially, it proactively takes the uncertainties associated with surrounding vehicles (SVs) into consideration, which encompass those stemming from the driver's intention, interactive behaviors, and the varying number of SVs. Intermediate decision variables are introduced to enable the high-level RL policy to provide an interaction-aware reference, for guiding low-level model predictive control (MPC) and further enhancing the generalization ability of the proposed framework. By leveraging the structured nature of self-driving at unsignalized intersections, the training problem of the RL policy is modeled as a bilevel curriculum learning task, which is addressed by the proposed Exp3.S-based BiMAB algorithm. It is noteworthy that the training curricula are dynamically adjusted, thereby facilitating the sample efficiency of the RL training process. Comparative experiments are conducted in the high-fidelity CARLA simulator, and the results indicate that our approach achieves superior performance compared to all baseline methods. Furthermore, experimental results in two new urban driving scenarios clearly demonstrate the commendable generalization performance of the proposed method.