The open racing car simulator (TORCS [14]), is a modern, modular, highlyportable multi-player, multi-agent car simulator. Its high degree of modularity and portability render it ideal for artificial intelligence research. Indeed, a number of research-oriented competitions and papers have already appeared that make use of the TORCS engine. The purpose of this document is to introduce the structure of TORCS to the general artificial intelligence and machine learning community and explain how it is possible to tests agents on the platform. TORCS can be used to develop artificially intelligent (AI) agents for a variety of problems. At the car level, new simulation modules can be developed, which include intelligent control systems for various car components. At the driver level, a low-level API gives detailed (but only partial) access to the simulation state. This could be used to develop anything from mid-level control systems to complex driving agents that find optimal racing lines, react successfully in unexpected situations and make good tactical race decisions. Finally, for researchers that like a challenge and are also interested in visual processing, a 3d projection interface is available.
[1]
Alex M. Andrew,et al.
Reinforcement Learning: : An Introduction
,
1998
.
[2]
Richard S. Sutton,et al.
Reinforcement Learning: An Introduction
,
1998,
IEEE Trans. Neural Networks.
[3]
Sylvain Gelly,et al.
Exploration exploitation in Go: UCT for Monte-Carlo Go
,
2006,
NIPS 2006.
[4]
Daniela Gorski Trevisan,et al.
Multimodal focus attention and stress detection and feedback in an augmented driver simulator
,
2007,
Personal and Ubiquitous Computing.