Spatiotemporal Receding Horizon Control with Proactive Interaction Towards Autonomous Driving in Dense Traffic

TL;DR

The paper tackles safe, high-performance autonomous driving in dense traffic by integrating a spatiotemporal receding horizon control (ST-RHC) framework with a proactive spatiotemporal safety barrier. It formulates a direct multiple shooting nonlinear program that jointly optimizes task goals, safety, and energy efficiency, solved in real time with warm-started SQP. Key contributions include the computationally efficient ST-RHC scheme, the differentiable spatiotemporal safety barrier with temporal attention, and extensive validation on synthetic IDM and real-world NGSIM data showing improved safety, accuracy, and efficiency over state-of-the-art baselines. The approach demonstrates robust real-time performance for horizons longer than 5 seconds, enabling proactive obstacle avoidance in multi-modal dense traffic and offering a solid foundation for deployment and extension to perception uncertainties in autonomous driving.

Abstract

In dense traffic scenarios, ensuring safety while keeping high task performance for autonomous driving is a critical challenge. To address this problem, this paper proposes a computationally-efficient spatiotemporal receding horizon control (ST-RHC) scheme to generate a safe, dynamically feasible, energy-efficient trajectory in control space, where different driving tasks in dense traffic can be achieved with high accuracy and safety in real time. In particular, an embodied spatiotemporal safety barrier module considering proactive interactions is devised to mitigate the effects of inaccuracies resulting from the trajectory prediction of other vehicles. Subsequently, the motion planning and control problem is formulated as a constrained nonlinear optimization problem, which favorably facilitates the effective use of off-the-shelf optimization solvers in conjunction with multiple shooting. The effectiveness of the proposed ST-RHC scheme is demonstrated through comprehensive comparisons with state-of-the-art algorithms on synthetic and real-world traffic datasets under dense traffic, and the attendant outcome of superior performance in terms of accuracy, efficiency and safety is achieved.

Figures (13)

• Figure 1: Illustration of the two edge collision cases between the EV (shown in red) and two surrounding HVs (shown in green and blue). The shadow color represents the anticipated motion of each vehicle.
• Figure 2: Top: illustration of the planned trajectory for the overtaking task, with the EV shown in red, and perceived and unperceived HVs depicted in orange and blue, respectively. Bottom: dynamic change in the heading angle $\varphi$. The heading angle decreases to a tiny value near the end of the planning horizon to stably rejoin its target racing lane.
• Figure 3: Comparison of position, velocity, acceleration, and heading angle profiles when executing an overtaking task using IDM dataset for surrounding HVs' motion. The similarities in the trajectory and heading angle profiles indicate how the EV attempts to adjust its heading angle to avoid a collision with SVs.
• Figure 4: Illustration of the EV's trajectory over the prediction horizon ($T = 5\,\text{s}$). The EV, represented by an ellipse, accelerates to overtake a front vehicle exhibiting multi-modal behaviors in four phases. Each phase is indicated by a unique colored rectangle, with the planned trajectory depicted as a blue dashed line. The red arrow denotes the current velocity vector of the AV. The text in each rectangle denotes the current velocity of each vehicle.
• Figure 5: Comparison of optimization time evolution for four algorithms in overtaking tasks with a prediction length of $N = 50$ in an adaptive cruise scenario, where the IDM governs the motion of surrounding HVs.

Theorems & Definitions (5)

• Definition 1
• Remark 1
• Remark 2
• Remark 3
• Remark 4