Presentation on theme: "The AGILO Autonomous Robot Soccer Team: Computational Principles, Experiences, and Perspectives Michael Beetz, Sebastian Buck, Robert Hanek, Thorsten Schmitt,"— Presentation transcript:
The AGILO Autonomous Robot Soccer Team: Computational Principles, Experiences, and Perspectives Michael Beetz, Sebastian Buck, Robert Hanek, Thorsten Schmitt, and Bernd Radig Munich University of Technology In Procs. of the First International Conference on Autonomous Agents and Multi-agent Systems Presented By: Jonatan Gomez
Outline Introduction Environment Sensors and Perceptions Drives and Goals Action-Selection Mechanisms (Control) Conclusions References
Introduction This paper describes: The computational model underlying the AGILO autonomous robot soccer team The AGILO implementation Some experience with it.
Introduction In robot soccer (mid-size league): Two teams of autonomous robots play soccer against each other. Each team has four members - one goal keeper and three field players. The soccer field is 4 * 9 meters big surrounded by walls.
Introduction Skillful play requires the robots: Recognize objects, such as other robots, field lines, and goals. Recognize entire game situations Collaborate by coordinating and synchronizing their actions to achieve their objectives.
Introduction The AGILO robot controllers employ: Game state estimation Situated action selection Playbook execution
Environment A soccer field with the following characteristics: 4 * 9 meters big Surrounded by walls Field lines 8 autonomous robots 2 Goals 1 ball
Drives and Goals Drives: Ultimate: Win the Game. Maximal: Score in the other team goal. Maximal: Do not allow to the other team score in its goal. Research: Show some kind of cooperative and collective behavior.
Drives and Goals Goals (Tasks): Intention of the AGILO robot team to perform a certain actions. Shoot the ball into the goal Dribble the ball towards the goal Look for the ball Block the way to the goal Get the ball … Each goal has an associated priority
Action-Selection Mechanisms Action-Selection (and execution) is constrained by: Goals being achievable only if certain conditions hold (eg, the robot has the ball) A robot is able to execute only one action at the same time
Action-Selection Mechanisms Situated Action-Selection: A goal assignment is a list of goals that an individual robot can perform (according to the goals priority and cost). An order over the goal assignments is imposed in order to determine the goals to be performed by a robot.
Action-Selection Mechanisms Situated Action-Selection: A is better than B if there is a goal in B that has lower priority that all the ones in A or they achieve the same goals but there exists a goal t in A such that all goals with higher priority are performed at least at fast as in B and t is achieved faster by A than by B
Action-Selection Mechanisms Situated Action-Selection: The goal (task) cost estimator perform three steps: Selection of the multi-robot navigation method that matches the game state best Computing a path in the context of the navigation paths of the team mates The proposed path is decomposed into a sequence of simpler navigation tasks for which the time cost can be accurately predicted using a neural network.
Situated Action-Selection: Choosing actions that have the highest expected utility in the respective situation Does not take into account a strategic assessment of alternative actions. In general is a limited temporal horizon.
Action-Selection Mechanisms Plan Based Control: Improve the robot soccer team by adding the capability of learning and execute soccer plays. Soccer plays are properly synchronized, cooperative macro actions than can be executed in certain game contexts and have, in these contexts, a high success rate.
Action-Selection Mechanisms Plan Based Control: A robot soccer playbook, a library of plan schema data that specify how to perform individual team plays, is added to each robot. The plans are triggered by opportunities, for example, the opponent team leaving a side open. The plays specify highly reactive, conditional, and properly synchronized behavior for the individuals players of the team.
Conclusions This paper described and discussed the control software of the AGILO autonomous robot soccer team. The AGILO teams employs sophisticated state estimation and control techniques, including experience-based learning and plan-based control mechanisms.
References M. Beetz. Structured Reactive Controllers. Journal of Autonomous Agents and Multi-Agent Systems, 4:25-55, March/June 2001. Sebastian Buck, Michael Beetz, and Thorsten Schmitt. Planning and Executing Joint Navigation Tasks in Autonomous Robot Soccer citeseer.nj.nec.com/445873.html Sebastian Buck and Michael Beetz and Thorsten Schmitt. M-ROSE: A Multi Robot Simulation Environment for Learning Cooperative Behavior citeseer.nj.nec.com/buck02mrose.html http://www.cs.kuleuven.ac.be/~nico/demo/pages/rcvi deo.html