Publications

This paper proposes a reward-driven automated curriculum reinforcement learning approach for interaction-aware self-driving at unsignalized intersections.

This paper proposes a novel learning-based MPC framework that incorporates full-state references and regulatory factors which can modulate the relative importance of each cost term within the cost functions.

A computationally-efficient ST-RHC scheme is proposed for autonomous driving, which leverages the multiple shooting method to improve computational efficiency and numerical stability, enabling accurate accomplishment of complex tasks in dense traffic scenarios in real time.

An interaction-aware RD-ACPPO framework is proposed for self-driving tasks at unsignalized intersections. Particularly, an adaptive curriculum learning technique is presented to accommodate a progressively increasing difficulty of learning tasks in autonomous driving.

This paper proposes a curriculum proximal policy optimization (CPPO) framework with stage-decaying clipping for unsignalized intersections.