P. Ritthipravat, T. Maneewarn, and D. Laowattana (Thailand)
Robot formation, Fuzzy logic controller, Q-learning, Formation sensitivity measure, Dynamic sub-goals generation.
Maintaining robot formation while moving through dynamic environment is one of the greatest challenges in robotic soccer. In this research, basic behavior of each robot is generated. A fuzzy logic controller is used to generate two basic robot behaviors: moving toward the predetermined sub-goal and avoiding moving obstacles. Two robots are set to make a line formation and move toward the ball, then to the goal. Formation sensitivity measure is introduced in order to evaluate the robustness of robot formation. High sensitivity means such formation is easy to be broken. In this research, sub-goals have to be dynamically determined. Using the Q-learning algorithm, dynamic sub-goals are generated for the fuzzy controller such that team of two robots can maintain their formation and move through moving obstacles toward their goal successfully. Sensitivity of formation is used as a reward/penalty value of learning algorithm. Simulation results show that, with this method, robots can move through moving obstacles toward the desired goal and maintain their formation successfully via generated sub goals.
Important Links:
Go Back
