KLOTSKI: Towards Consensus Enabled Collaborative Vehicles in Intelligent Transportation

With the fast development of Vehicle to Everything (V2X) and vehicular automation, vehicles communicate with infrastructure and each other to enable cooperative decision-making, fostering advancements in autonomous driving and intelligent transportation systems. However, coordinating vehicular decisions in general transportation scenarios like lane change and intersection traffic has not been systematically explored so far. To address this gap, this paper investigates cooperated vehicle systems that utilize V2X technology to align decision-making in transportation. To this end, we identify the central challenge in vehicular coordination as the “consensus” problem in transportation, allowing vehicles to reach an agreement on a global schedule and then safely execute their actions. Furthermore, we propose <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Klotski</small>, a framework to enable vehicles to achieve consensus by considering challenges such as latency and jitter of V2X communication, vehicle faults, and dynamic participants. To better address these challenges, <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Klotski</small> framework employs a modular design comprising three core components: the Membership Management Module, Intelligent Decision Module, and Coordination Module. Furthermore, we illustrate the practical application of <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Klotski</small> in two real transportation scenarios while highlighting its desired properties through security analysis. We build a prototype of <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Klotski</small> and evaluate its performance based on discrete event simulations. The evaluation results show that, without considering crashes, our scheme can reach consensus 27 times per minute, while participant crashes increase the consensus time by an average of 80.5%.