Key Technologies Of Multi-robot Coordination And Cooperation In Adversarial Environment

Multi-robot coordination and cooperation are hot issues now. It is mainly on the problems of multi-robot coordination and cooperation in adversarial environment. In adversarial games, there is conflict of interest among players. In general, players usually take the action that seems to prevent their opponents from victory. The demands in military field and robots game make it meaningful to research on multi-robot coordination and cooperation in adversarial environment.It uses middle size soccer robot as the research platform. Middle size soccer robot game is a procedure that each robot configures itself in space and time space according to the adversarial situation and executes certain behavior such as dribbling and kicking. A survey of the state-of-art of middle size soccer robot is presented. It also presents a review of key technologies of multi-robot coordination and cooperation, including the architectures of robot intelligent systems, world modeling, behavior decision-making and cooperative learning, etc.There are two preconditions of perfect teamwork. One condition is that each robot has a consistent global world modeling; the other is that for each kind situation, there are conventions among robots on strategic and tactic, coordination and cooperation mechanism and behavior decision-making beforehand. Thus, it is mainly on the issues of cooperative world perception, cooperative mechanism and behavior decision-making. The main achievements of the paper are as follows: Because it is very difficult for a today robot to establish a precise global world model just with its own sensor information, a method to establish precise global world model with multi-robot cooperation is presented. At first, each robot localizes itself with vision and odometer information fusion; then obtains the absolute coordination values of objects seen; with the presumption that the self-localization of each robot is trustable, identifies the identifications of all objects on the field and obtain their positions; thus, global world models are established. Identifications and positions of objects'are obtained with the analysis method of congregating.Based on the practice of RoboCup, multi-robot coordination and cooperation technologies for two levels of world modeling are discussed. One level of world modeling is that each robot can localize itself, obtain teammates'poses and approximate opponents'positions. It is the current world modeling level of robots. The other level of world modeling is that each robot can not only obtain poses of all objects on the field, but also obtain the motion estimation of opponents and ball.According to the current world modeling, a kind of decision-making procedure is presented as follows: at first, each robot sorts the current adversarial situation and estimates the current task of the team; then following certain strategy and rule, robot evaluates the suitability of roles for all teammates and estimates all teammates'rational behavior; at last, robot selects a rational action to execute. To implement it, the possible situations and the corresponding team tasks should be analyzed beforehand. The rules of role assignment and behavior making method should also be designed beforehand. It should be noticed that roles assignment should be smooth and consistency during the shift procedure from one situation to another situation. A cooperation mechanism integrated with dynamic role assignment and fixed region defence is presented. It uses the formation-based team organizing mechanism. Team tactics are implemented by certain formation configuration, attack route and shift of attack route. Because of the requirements from attack and defence in soccer robot game, a team organization and formation control method with two leaders is presented. The passive defensive formation is organized by taking goalie or rear guard as leader according to situation. The active attack formation is organized by taking the current critical robot as the leader. According to the feature of role division method applied in soccer robot game, it presents an idea that a global coordination and cooperation is implemented by local coordination and cooperation. It proposes a plan method that according to the emergency degree, local coordination and cooperation is put in a priority queue.A short-term predictive adversarial plan method based on quantified analysis is presented and analyzed in Chapter 4. The short-term predictive cooperative adversarial plan method adopts the assumption that robots do not know whether their opponents are rational or not. The basic procedure of short-term predictive method consists of three steps: at the first step, works out possible states of all players at certain short-term futures; then evaluates the expected payoffs of possible actions; at last, select an action to execute based on the decision-tree of expected payoffs.Because that robot likes to execute certain action in certain situation, the concept of action preference is introduced. Integrated with the adversarial plan method based on short-term prediction, a self-adaptive adversarial plan method is proposed. With game going on, the estimation of opponents'action preferences is adjusted by observation. Thus the estimation of opponents'behaviors will be better and the action decision would be more and more preferable. Furthermore, to make the design of complex action and cooperation to be more convenient, a primary design of graphical action and formation design system is presented.The theories and algorithms presented in the paper are either tested with real middle-size soccer robots or verified with simulations. The results of experiments illustrate that they are useful.