Virtual Chaotic Traffic Simulation Gaurav Chaurasia * B. Radhika Selvamani Nithi Gupta Subodh Kumar † Department of Computer Science & Engineering IIT Delhi New Delhi 110016 INDIA ABSTRACT This paper presents a novel traffic simulation scheme capable of modeling most forms of urban, chaotic traffic. Different from other lane-based or following-based approaches, ours models traffic as a large navigational problem in an agent based simulation context. While this generalization makes the traffic more reflective of certain scenarios, it also leads to some complexity that we address. It is able to handle dense traffic and selects for each car, independently, the op- timal velocity and acceleration to find a path through a fast evolving obstacle network. The selection of parameters at each simulation step is posed as an optimization problem ensuring smooth motion subject to car kinematics. In addi- tion to overtaking, the approach is efficiently able to handle hard cases like behavior at traffic lights and turning. We demonstrate our simulation at real-time rates using average computing resources. 1. INTRODUCTION Traffic simulation has many applications ranging from ur- ban planning to computer games to movies to learning about driver behavior. It follows naturally that the area boasts a rich body of literature. Much of this literature is in the con- text of streamlined lane-based traffic flow, as is common in many parts of the world. As a result, many of the ideas are inapplicable to chaotic traffic scenarios as is prevalent in many other parts of the world, where sometimes lane markings may not even exist. We target chaotic traffic sim- ulation. In particular, our goal is to visualize low and high traffic virtual environments: we focus on accurate display of vehicle behavior. Our design allows the display and sim- ulation loops to run asynchronously. This allows the scene to be displayed at a high frequency, even if the simulation frequency is somewhat lower. Traffic simulation is often categorized into macro and mi- cro simulation (with some examples of hybrids) [20]. Micro- * Currently at REVES/INRIA Sophia Antipolis † Contact Email:subodh@cse.iitd.ac.in Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Copyright 2010 ACM 978-1-4503-0060-5/10/12 ...$10.00. simulation schemes simulate each vehicle, while macro-simul- ators are concerned mainly with statistics and volumes of traffic flows and bottlenecks. We focus on micro-simulation, for one of our main applications is computer games – one can imagine racing the streets of Bangalore with other gamers, all while “normal” traffic is in full flow. Macro-simulation does not lend itself to such rendering. Furthermore, response to developing traffic situations (like bottlenecks and jams) can be better handled by micro-simulation. We would like our simulation to be realistic enough for urban planning, or for games and movies in the context of, say, urban warfare or criminal pursuit. Our immediate goal is for the physical simulation as well as the resulting graphics to look realistic. We leave a detailed physical validation of the simulation for later. We present a parameterizable generic traffic model with few hard-coded assumptions. We neither assume strict lane based driving nor pre-define overtaking trajectories [18], which may apply only to sparse highway traffic. Our technique is able to simulate chaotic urban traffic, and yet converge to more structured movement in low traffic density or low driver aggression levels. 2. PREVIOUS WORK Our problem is primarily one of simulating behavior of multiple agents. Continuum based multi-agent approaches [14] model the system with a set of partial differential equa- tions. Each agent in the system is associated with a force or potential field, and the proximity of the agent to other agents or obstacles is captured in terms of high energy states. The optimal path is one that assures a low energy state to the agent. The behavior of all the agents is the solution of this global system of equations, and therefore such a system is centralized in nature. The major advantage of this ap- proach is that it can capture arbitrary agent motion. And the major challenge is that solving the system of equations can be numerically unstable, especially in situations where the functions are not continuous or differentiable. More im- portantly, although the results seem plausible at a macro level, individual agent behavior may seem artificial. Other centralized approaches also suffer from similar unre- alistic behavior. The fact is that real drivers are autonomous and are not always able to find the best paths. Indeed, traf- fic involves a complex and dynamic interaction of multiple independent decision makers in the presence of limited re- sources (the road). We pursue a decentralized approach, which seems to more naturally capture such complex deci- sion making. Each agent is autonomous and decides its path