A low-cost Framework for Decentralized Autonomous Intersection Management
Rugved Katole, Arpita Sinha
TL;DR
The paper tackles unsignaled intersections under low-to-moderate traffic by introducing a low-cost, decentralized framework that uses an offline harmony matrix to identify non-conflicting vehicle maneuvers. It constructs a graph from the harmony matrix and solves a maximal clique to select the best crossing set, with lane-priority tie-breaking and a red-yellow-green zone scheme to ensure safety and deadlock-free operation. Validated in SUMO with Poisson arrivals across 3-, 4-, and 5-way intersections, the approach delivers lower waiting times than fixed-time and adaptive signals and performs comparably to V2I-based strategies at densities below 500 PCUs/hr/lane. The framework reduces infra-structural costs while maintaining throughput, and is extensible to higher-order intersections and mixed autonomy in future work.
Abstract
This paper addresses the traffic management problem for autonomous vehicles at intersections without traffic signals. In the current system, a road junction has no traffic signals when the traffic volume is low to medium. Installing infrastructure at each unsignalled crossing to coordinate autonomous cars can be formidable. We propose a novel low-cost solution strategy where the vehicles use a harmony matrix to find the best possible combination of the cars to cross the intersection without any crashes. The harmony matrix defines the connection between different vehicle maneuvers and is queried online for intersection management. We maximize the throughput of the intersection by solving a maximal clique problem formulated based on the vehicles present at the intersection. The proposed algorithm relies on the intent perceived by the autonomous vehicles. We compare our work with a communication-based strategy that uses V2I communication protocols, and through extensive simulation, we showed that our algorithm is comparable when the traffic volume is less than 500 PCUs/hr/lane.