1 / 19

Torcs Simulator

Torcs Simulator. Presented by Galina Volkinshtein and Evgenia Dubrovsky. Overview. Torcs Motivation Optimization Algorithm Results and Comparison. Torcs. Car Setup Optimization Competition @ EvoStar 2010. The purpose is to find the best car setup.

carya
Download Presentation

Torcs Simulator

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Torcs Simulator Presented by Galina Volkinshtein and Evgenia Dubrovsky

  2. Overview • Torcs • Motivation • Optimization Algorithm • Results and Comparison

  3. Torcs • Car Setup Optimization Competition @ EvoStar 2010. • The purpose is to find the best car setup. • The contest is divided into an optimization phase and an evaluation phase. • During the optimization phase, the optimization algorithm will be applied to search for the best parameter setting. • During the evaluation phase, the best solution will be scored according to the distance covered in a fixed amount of game time.

  4. Overview • Torcs • Motivation • Optimization Algorithm • Results and Comparison

  5. Motivation • The winner of the competition evostar2010 in April -Jorge Muñoz used a MOEA. • MOEA - Multiobjective evolutionary algorithm. • MOEAs: • Aggregation based – non-dominated solutions are obtained by a weighted sum of the individual objective functions. • Dominance based – use the dominance relation as a measure of the fitness of each individual.

  6. Overview • Torcs • Motivation • Optimization Algorithm • Results and Comparison

  7. NSGA-IIIntroduction. • The ranking-based evolutionary algorithm NSGA-II combines elitism and a mechanism to distribute the solutions as much as possible. • Multiobjective optimization elitism requires that some portion of the non-dominated solutions will survive.

  8. NSGA-IIIntroduction (cont'd). • NSGA-II is based on dominance count. • Multiobjective optimization populations can search many local optima so a finite population tends to settle on a single good optimum, even if other equivalent optima exist. • Special mechanisms are required to prevent this occurring. • Niche induction methods promote the simultaneous sampling of several different optima by favoring diversity in the population.

  9. NSGA-IIIntroduction (cont'd). • Individuals close to one another mutually decrease each other's fitness. • Isolated individuals are given a greater chance of reproducing, favoring diversification.

  10. NSGA-IIComponents. • Dominance. • Only non-dominated solutions are kept. • Crowding. • Density less crowded regions are preferred to crowded regions.

  11. NSGA-IIFlow. • NSGA-II classifies a population in several classes which are called fronts. • The number of classes varies from generation to generation and the members in each class are equivalent. • That is, it cannot be stated which individual is better. • This classification which is called non-dominated sorting is implemented as follows.

  12. NSGA-IINon-dominated Sorting. • All non-dominated individuals are classified into one category and assigned a dummy fitness value or rank. • These classified individuals are ignored and from the remaining members of the population the non-dominated individuals are selected for forming the next layer. • This process continues until all members are classified. • Individuals of the first layer have the highest fitness while members of the last layer are assigned the smallest fitness. • All individuals from the first layer produce more copies in the next generation.

  13. NSGA-IICrowding distance. • An estimation of the density of solutions surrounding each member is calculated using the crowding distance: • The population is sorted in ascending order. • The solutions with the smallest and largest value are assigned a very large distance estimate to guarantee that they will be selected in the next generation. • All other solutions are assigned a distance value equal to the absolute difference in the function values of two adjacent solutions.

  14. NSGA-IIElitism. • The elitism is used by combining together the population of children Q_t and the parent population P_t at generation t together. • A non-dominated sorting is applied and a new population is formed. • A population of children Q_t+1 from P_t+1 is formed using a binary crowded tournament selection, crossover and mutation.

  15. Overview • Torcs • Motivation • Optimization Algorithm • Results and Comparison

  16. Results and Comparison E-track instead of Poli-track

  17. Results and Comparison

  18. The End Any Questions ?

  19. The End Thank you ;)‏

More Related