Application of reinforcement learning on self-tuning PID controller for soccer robot multi-agent system

In the soccer robot game, quickly and accurately move is a very necessary. Proportional-Integral-Derivative (PID) controllers are applied to overcome it. However, to obtain the optimal control of robots, required tuning PID of parameters (Kp, Ki, and Kd) that not easy. In this research used reinforcement learning algorithm with Q-Learning method to determine the parameters in the process of self-tuning PID control. To quantify the results that obtained, it would require a comparison between the use of RL algorithms on self-tuning PID with PID conventional usage by tuning using Ziegler-Nichols oscillation method, and compared to controls that have been implemented in the YSR-A Yujin Robotics. Test was conducted by moving the 5 robot with combination 3 robots of red and 2 white robot toward a particular point and to play with a certain angle. From the test results is known that the values obtained using the RL algorithm parameters PID control output which produces the most stable and have 1.875 faster response than using the methods applied YSR-A Yujin Robotis and have 1.5 faster response compared by using the Ziegler-Nichols oscillation method.