Generalized Coordination of Partially Cooperative Urban Traffic

TL;DR

A novel cooperative maneuver planning approach that is generalized to various situations found in urban traffic and based on an optimization approach accompanied by an efficient heuristic method for high-load scenarios is presented.

Abstract

Vehicle-to-anything connectivity, especially for autonomous vehicles, promises to increase passenger comfort and safety of road traffic, for example, by sharing perception and driving intention. Cooperative maneuver planning uses connectivity to enhance traffic efficiency, which has, so far, been mainly considered for automated intersection management. In this article, we present a novel cooperative maneuver planning approach that is generalized to various situations found in urban traffic. Our framework handles challenging mixed traffic, that is, traffic comprising both cooperative connected vehicles and other vehicles at any distribution. Our solution is based on an optimization approach accompanied by an efficient heuristic method for high-load scenarios. We extensively evaluate the proposed planer in a distinctly realistic simulation framework and show significant efficiency gains already at a cooperation rate of 40%. Traffic throughput increases, while the average waiting time and the number of stopped vehicles are reduced, without impacting traffic safety.

Figures (7)

• Figure 1: Four of the 13 simulated real-world scenarios. They comprise five intersections with right-before-left regulation (a) or main and side roads (b), seven roundabouts (c), and a road narrowing (d). Green/yellow/red vehicles are coordinated CAVs with early/medium/late entrance time; uncoordinated CAVs are blue; gray vehicles are HDVs. Dark red areas denote conflict zones.
• Figure 2: System overview. Parts handled in this work are marked in red, dashed lines denote V2X communication. The two modules on the left are deployed on an edge server close to the road users and infrastructure on the right.
• Figure 3: Prediction accuracy per vehicle of our $\mathrm{MLP}_\mathrm{acc}$ vs. IDM on the inD and openDD datasets. Our gap acceptance model $\mathrm{MLP}_\mathrm{gap}$ was used for both.
• Figure 4: Results of 135 simulations at four intersections with a main road.
• Figure 5: Results of 43 simulated scenarios at one right-before-left intersection.