TORCS Training Interface : uma ferramenta auxiliar ao desenvolvimento de pilotos do TORCS

The ine cient manner in which drivers are tested and developed for the racing game and simulator TORCS is a relevant problem because of the limitations imposed over projects of development of drivers, i.e., algorithms that determine the behavior of cars that are not controlled by human players. Because this software has a role of benchmark for di erent techniques of Arti cial Intelligence, it is important to work on mitigating this problem. The TORCS Training Interface was developed, a tool that o ers automatizations in order to improve the e ciency of simulation calls and return more complete data both of which are important for the necessary evaluations that have as a goal estimating the tness of drivers. Results of the comparative tests performed indicate that the use of the tool is a viable alternative to the approaches seen in the literature, presenting advantages that can make it the most tting to processes that are similar to the ones considered here.