Approximately Orchestrated Routing and Transportation Analyzer: Large-scale Traffic Simulation for Autonomous Vehicles Dustin Carlino, Mike Depinet, Piyush Khandelwal, and Peter Stone Department of Computer Science The University of Texas at Austin Austin, TX 78712 {dcarlino,msd775,piyushk,pstone}@cs.utexas.edu Abstract— Autonomous vehicles have seen great advance- ments in recent years, and such vehicles are now closer than ever to being commercially available. The advent of driverless cars provides opportunities for optimizing traffic in ways not possible before. This paper introduces an open source multiagent microscopic traffic simulator called AORTA, which stands for Approximately Orchestrated Routing and Transporta- tion Analyzer, designed for optimizing autonomous traffic at a city-wide scale. AORTA creates scale simulations of the real world by generating maps using publicly available road data from OpenStreetMap (OSM). This allows simulations to be set up through AORTA for a desired region anywhere in the world in a matter of minutes. AORTA allows for traffic optimization by creating intelligent behaviors for individual driver agents and intersection policies to be followed by these agents. These behaviors and policies define how agents interact with one another, control when they cross intersections, and route agents to their destination. This paper demonstrates a simple application using AORTA through an experiment testing intersection policies at a city-wide scale. I. INTRODUCTION Autonomous vehicle technology has made tremendous progress in the last decade. In 2007, six of the competing teams completed the 96 km course set for the DARPA Urban Challenge [1]. They did so while obeying the traffic laws fol- lowed by human drivers, navigating along with other moving vehicles, and following correct intersection precedence order. Since then, Google’s driverless cars have clocked more than 250,000 km on public roads in urban California, USA [2]. In 2010, researchers from the University of Parma successfully completed an autonomous intercontinental run from Parma, Italy to Shanghai, China [3]. The successful completion of all these milestones suggests that autonomous cars are here to stay, and are ever closer to becoming commercially available. With the arrival of autonomous cars, it also becomes possible to optimize traffic in ways not possible for human drivers. This paper introduces an open source multigent micro- scopic traffic simulator called AORTA, which stands for Ap- proximately Orchestrated Routing and Transportation Ana- lyzer. AORTA is a platform designed for testing autonomous vehicle behaviors and intersection policies. Autonomous vehicles, termed agents, use behaviors to interact with one another, follow intersection policies, and decide on both long term and short term actions. Intersection policies designate when it is safe for an agent to cross an intersection. AORTA’s Fig. 1: Visualizing autonomous agents in downtown Austin, Texas, with AORTA’s UI goal is to allow for the definition of new agent behaviors and intersection policies to optimize autonomous traffic. Additionally, by assigning human-like behaviors such as the “car following model” [4] to agents and traffic-signal-like policies to intersections, AORTA could potentially simulate human traffic. Like any other microscopic traffic simulator, AORTA needs maps to run simulations. One of AORTA’s key features is that it generates maps using real road data available from OpenStreetMap (OSM) [5], a moderated, user-editable interface for world maps. A map for any desired city in the world can be downloaded, which is then parsed by AORTA to set up a scale simulation of the real world in a few minutes. AORTA is available open-source and is easily extensible, 1 making it easy for users to test out a number of agent behaviors and intersection policies in a short time span. This paper explores the state-of-the-art in traffic simulators in Section II, followed by a description of AORTA’s architec- ture in section III. Use of OSM data in AORTA is explained in Section IV, along with a description of the simulator in Section V. Section VI demonstrates an application built on top of AORTA and presents a simple experiment evaluating different intersection policies in a large scale scenario. The paper then concludes with a discussion on future work. II. RELATED WORK Computational processing power has made excellent ad- vancements in the last two decades. Parallel computing and the use of GPUs have enabled microscopic models of traffic 1 Code available at http://code.google.com/p/road-rage In Proceedings of the 15th IEEE Intelligent Transportation Systems Conference (ITSC 2012), Anchorage, Alaska, USA, September 2012