Development of Push-Recovery control system for humanoid robots using deep reinforcement learning

Abstract

This paper focuses on the push-recovery problem of bipedal humanoid robots affected by external forces and pushes. Since they are structurally unstable, balance is the most important problem in humanoid robots. Our purpose is to design and implement a completely independent push-recovery control system that can imitate the actions of a human. For humanoid robots to be able to stay in balance while standing or walking, and to prevent balance disorders that may be caused by external forces, an active balance control has been presented. Push-recovery controllers consist of three strategies: ankle strategy, hip strategy, and step strategy. These strategies are biomechanical responses that people show in cases of balance disorder. In our application, both simulation and real-world tests have been performed. The simulation tests of the study were carried out with 3D models in the Webots environment. Real-world tests were performed on the Robotis-OP2 humanoid robot. The gyroscope, accelerometer and motor data from the sensors in our robot were recorded and external pushing force was applied to the robot. The balance of the robot was achieved by using the recorded data and the ankle strategy. To make the robot completely autonomous, Deep Q Network (DQN) and Double Deep Q Network (DDQN) methods from Deep Reinforcement Learning (DPL) algorithms have been applied. The results obtained with the DDQN algorithm yielded 21.03% more successful results compared to the DQN algorithm. The results obtained in the real environment tests showed parallelism to the simulation results.

Keywords

• Deep reinforcement learning
• deep q network(DQN)
• double deep q network(DDQN)
• Humanoid robot
• Robotis op2
• Push-recovery