הנכם מוזמנים לסמינר של מאיר הראל - דוקטורנט - חיפוש אסטרטגיות לבקרת רובוט עבור שימושים בסביבה עם יריב תחת ריבוי קריטריונים
School of Mechanical Engineering Seminar
Monday, November 23, 2020 at 14:00
Multi-criteria Search of Robot Control Strategies for Applications in an Environment with an Adversary
Ph.D. student under to supervision of Dr. Amiram Moshaiov
Multi-Objective Games (MOGs) are games in which each player has more than one objective to accomplish. Given the multiple objectives, each player may experience a self-conflict about its objective preferences. In recent studies, a novel solution approach to non-cooperative MOGs has been suggested which is termed the rationalizability solution concept. This approach assumes that the players of the MOG are undecided about their objective preferences. Employing the rationalizability approach provides each of the players with a Set of Rationalizable Strategies (SRS). This set exposes the performance tradeoffs among the various rationalizable strategies.
Following the rationalizable solution concept, the current work suggests a modification to the rationalizability solution concept and introduces an approach to reduce the SRS. The suggested modification, which incorporates subjective preferences of the objectives by the players, is based on some Multi-Criteria Decision-Analysis (MCDA) ideas. An evolutionary algorithm, which is based on the introduced modification, is developed. This algorithm results in a reduced SRS which is termed the Set of Preferred Strategies (SPS). The proposed modification is realized by the introduction of auxiliary criteria into the evolutionary search, which could reduce the computational efforts and supports the decision-making.
In addition to modifying the rationalizability approach and the development of the associated algorithms, this work investigates the implementation of the proposed ideas to real-life robotic/dynamic MOGs. For this purpose, a real-life aerial MOG is taken, which involves a navigator and a coalition between the navigation target and a missile that pursuits the navigator. To support solving the aforementioned MOG a novel analytical solution of the problem is introduced. The resulted sets of all possible strategies are analysed using full sorting according to the rationalizability concept. Next, the analytical solution undergoes a numerical modification using neural-networks. The proposed modification undergoes a search using the proposed evolutionary algorithm to produce rationalizable strategies to the original MOG problem. Finally, the resulting control strategies were analysed and examined for their performance characteristics under some changes to the initial conditions of the game.
https://zoom.us/j/96584758181?pwd=WC9PMXdsYzJ3NFdEN2Q5ZUtOZEVjdz09 The meeting will be recorded and made available on the School’s site.