Reinforcement learning and neural network-based artificial intelligence control algorithm for self-balancing quadruped robot.

This paper proposes an artificial intelligence (AI)-based new control algorithm for a self-balancing quadruped robot. A quadruped robot is a good example of a redundant degree-of-freedom (DOF) system and is designed for locomotion over extreme terrain conditions. Even though a relevant control algorithm exerts a great effect on the performance of the locomotion control of quadruped robots, controlling them is difficult and complex because of the redundant DOF and interlocked movement of their four legs. This paper presents an effective control algorithm that can replace the typical analysis-based control theory, including inverse kinematics, differential equations of motion, and governing equations, which is based on reinforcement learning (RL) and artificial neural network (ANN). RL generates the training data to train the ANN model, and the trained ANN model is finally used to control a quadruped robot. The proposed AI-based robot-control algorithm is validated experimentally using a customized test-bed and a self-balancing quadruped robot. The results show that the proposed method is a promising new control algorithm that can replace the mathematically incomprehensible robot-control system. [ABSTRACT FROM AUTHOR]