Harmonic opponent modeling and behavior structure for 3D soccer simulation agent

Hierarchical Behavior Structure for NAO Robot Utilizing Fuzzy Inference Systems

“…In this paper we implemented and testes two different fuzzy inference systems on the "ball-handled" behavior in the behavior structure we introduced in [8], lucasiewicz inference engine with largest maixa defuzzifier gained better results than product inference engine with maxima, average of centers and center of gravity defuzzifiers. Playing soccer in not only NAO robot but also in any humanoid robot can get better and better by using fuzzy logic on more and more behaviors and of course by combination of fuzzy logic with evolutionary algorithms.…”

“…In this skill in our previous work we assumed a threshold for a distance between an agent (NAO robot in 3D soccer simulator) and ball, the robot will approach to ball as its real distance to it become smaller than the threshold we assumed [8]. As the matter of fact, in this skill we want our agent to know whether or not the ball is in its controllable area, in this regard, we profited from fuzzy inference system which will be completely described in the next section.…”

“…Winning the game in robot soccer depends on many factors, some of the most important factors are: the ability of our agents which means, for instance, our agent's walking speed, the ability of opponent agents which means the same as our agents ability, one other important factor is how precisely our agent can guesstimate the ability of its opponent agents [9], in this regard we proposed harmonic opponent modeling in our previous work [8], another significant factor of winning is making decisions properly during the game, in kick-walk decision making behavior, the output of opponent modeling is used, another behavior in which a decision is made is ball-handled in which as we mentioned above, the robot can decide if the ball is in its controllable area or not, in this regard we benefited from fuzzy inference systems. Two different kinds of fuzzy inference systems in which one of them uses product fuzzy engine and the other uses lukasiewicz inference engine has been implemented, the product inference engine (PIE) uses mamdani product logic, algebraic product for t-norms and maximum operator for s-norms, we used three different kinds of maxima, average of centers, and center of gravity defuzzifiers to defuzzify the PIE output, the lukasiewicz inference engine (LIE) uses lukasiewicz logic, and minimum operator for t-norms, we used maxima defuzzifier with LIE, for each inference engine we will describe the formulization of the first rule in fuzzy rule base.…”

“…for instance, Loetzsch et al in Four-Legged league introduced the XML based tool for engineering the behavior of autonomous agents in complex and dynamic environments, called XABSL [3], which describes the behaviors of an agent with a set of finite state machines, the transition between states are modeled as decision trees, in Middlesized league, Utz et al proposed a framework called BAP [4] in which hierarchical, behavior-based control systems can be specified, or another example for Middle-Size league and of course for humanoid league is Dual Dynamics architecture developed by Jaeger and Christaller in which behaviors are specified as dynamical systems using differential equations [5], Hertzberg and Schonherr combined Dual Dynamics with planner and developed another framework which called DD & P [6], another example for behavior structure which is used in two leagues was introduced by Laue and Rofer [7] which is used in four legged and small-sized league. Introducing the 3D soccer simulation league and using humanoid robots to play soccer, new challenge of humanoid soccer serves as benchmark problem for artificial intelligence, standard robot of this league is NAO robot which has its own league in RoboCup competitions nowadays, it has 58cm height, 4.3 kg weight and 21-25 degrees of freedom, we proposed a hierarchical behavior structure for NAO robot in [8] which is based on finite state machine, in this paper we implement and compare two different fuzzy inference systems on the "Ball-handled" behavior in the mentioned behavior structure. In section 2, one can find a brief description of behavior structure and Ball-handled behavior, in section 3, two fuzzy inference systems has been explained, in section 4 we describe the implementation and experimental results, and finally on section 5, conclusion and future works has been mentioned.…”