Multiagent system architecture and method for group-oriented traffic coordination Jana G¨ ormer and J¨ org P. M¨ uller Department of Informatics, TU Clausthal, Germany Email: {Jana.Goermer|Joerg.Mueller}@tu-clausthal.de Abstract—Next-generation traffic management systems will make use of on-board intelligence and communication capabilities of vehicles and traffic infrastructure. In this paper, we investigate a multiagent approach allowing vehicle agents to form groups in order to co-ordinate their speed and lane choices. Our hypothesis is that a decentralized approach based on a co-operative driving method can contribute to higher and smoother traffic flow, leading to higher speeds and less delays. Our focus is on automated vehicle decision models. We develop a group-oriented driving method with vehicle agents that perceive their environment and exchange information. The paper proposes decentralized dynamic vehicle grouping algorithm, a conflict detection and global coordination method, and defines individual driving strategies for vehicles. For validation, we compare our method with a driving method implemented in the commercial traffic simulation platform AIMSUN. Experimental results indicate that group formation and group coordination methods can improveme traffic network throughput. MAS coordination, MAS cooperation, traffic vehicle grouping (key words) I. I NTRODUCTION Next-generation traffic systems will feature onboard intelli- gence and Car-to-X communication capability. This implies new control challenges and a shift from traditional hierarchical organization to a multiagent systems organization. But it also opens ample of new opportunities in terms of ad-hoc coordi- nation and co-operation of vehicles and infrastructure compo- nents in order to maximize throughput and avoid breakdowns. Our conception is that in future cyber-physical traffic and transport ecosystems, vehicles and infrastructure components will be equipped with software agents that make decisions autonomously 1 on behalf of traffic participants and control authorities. In this work, we develop an agent-based model and corresponding distributed algorithms that represent different types of traffic participants with local goals, differing capaci- ties (speed, acceleration, driving skills), and preferences (speed or lane). The driving behavior of a vehicle (e.g. acceleration, deceleration, or lane changing) is a function of its parameters and of the traffic regulations. However, the ability of a traffic participant to act according to its preferences is restricted by other participants: Often, fast vehicles are forced to drive slow This work was funded by the NTH Focused School for IT Ecosystems (www.it-ecosystems.org). 1 We acknowledge the role of the human-in-the-loop, but it is not the focus of this work. because overtaking or lane changing is not possible due to slower vehicles (speed conflict problem). Consequently, traffic congestion occurs and the travel times increase. In this work, we approach the problem illustrated in the scenario by proposing Group-oriented driving (GoD), a new autonomous co-operative driving method. In GoD autonomous vehicles are able to perceive their environment, communicate, form groups, and co-ordinate their behaviors in order to avoid and solve the conflict situations described above. We compare our method with a (non-cooperative) reference driving method implemented within the commercial traffic simulation platform AIMSUN. We give simulation experiments that suggest that the method can reduce travel times and delays, while the overall benefit of the approach depends on the structure of the overall vehicle population. The paper is organized as follows: After discussing related work in Section II, we give an overview of the GoD method and the architecture of the multiagent system (MAS) used in this work in Section III. Section IV describes the cooperation protocol between autonomous vehicle agents and presents the driving strategy of individual autonomous vehicle in the context of the cooperative method. In Section V a case study is used to evaluate the performance of GoD in terms of speed and travel time. II. RELATED WORK Methods of coordination and cooperation in MAS are con- sidered e.g. in [1]–[3]. Barrett et al. [4] examine models for ad- hoc agent teamwork for an abstract simplified domain without considering communication and sensory aspects. There are various multiagent grouping technologies applying general coordination and cooperation processes, e.g. [5], [6], but no solutions tailored for the traffic domain. Therefore, we provide a method for modeling, simulating, and analyzing various traffic scenarios with autonomous vehicles at micro- scopic level. On the run-time side, we use AIMSUN [7] as a traffic simulation system to model and simulate traffic scenarios, and integrate it with the JADE framework [8] for agent-based implementation. The resulting platform is “Agent- based Traffic Simulation System (ATSim)” [9]. We realize a solution to support global system throughput on a macro- level view while preserving decentralization and openness on the individual vehicle micro-level. Hence, the infrastructure elements of the traffic domain like traffic lights and/or vehicles 978-1-4673-1703-0/12/$31.00 ©2013 IEEE