Cooperative Strategies and Capture Zones in Multi-Player Pursuit-Evasion Games
Work towards PhD degree under the supervision of Assoc. Prof. Tal Shima
Department of Aerospace Engineering
Technion – Israel Institute of Technology
Cooperative strategies are becoming more and more popular with the continuing evolution and advancement in decision making capabilities of autonomous vehicles. Utilizing multiple agents to perform a given task can be beneficial even in cases when the goal is achievable by a single agent. With regard to interception engagements of an intelligent target, through shared information and coordinated actions the capability requirements and/or the number of required agents in order to capture the target may be relaxed and reduced, respectively. It is therefore a great point of interest, when analyzing such problems, to discern under what conditions the evader’s capture can be guaranteed (and whether or not these conditions are dependent on the number of pursuers and their capabilities) and which element of cooperation (if any) can help to achieve this guarantee.
We will begin the talk with an examination of the solution of a linear 1-on-1 interception engagement with multiple bounded control inputs, formulated as a two-person, zero-sum, pursuit-evasion game, from which we derive the conditions under which a single pursuer may guarantee the exact capture of a single (possibly) evading target, extending current known results. This will be followed by an analysis of a 2-on-1 pursuit-evasion game (a team of two pursuers vs. a single evader) in which we consider two different formulations: the first including a product-form cost function and the second using a guaranteed miss approach. Based on the latter, we will propose a cooperative pursuit strategy for a couple of “weak” pursuers and illustrate its performance through simulations. Finally, following some observations from the 2-on-1 game, we will present necessary and sufficient conditions by which a group of (possibly) cooperating pursuers may guarantee point capture of a single (possibly) evading target in a general n-on-1 linear engagement, showing that, with respect to guaranteeing point capture, the pursuers cannot actually benefit from cooperating with one another.
The talk will be given in English
Wed, 19-04-2017, 16:30 (Gathering at 16:00)Classroom 165, ground floor, Library, Aerospace Eng.
Light refreshments will be served before the lecture