Deep Reinforcement Learning (DRL) algorithms have been successfully used in a wide range of challenging control tasks for various robot environments. In this paper, we study a training procedure for a quadruped robot locomotion through DRL. We apply a...
Deep Reinforcement Learning (DRL) algorithms have been successfully used in a wide range of challenging control tasks for various robot environments. In this paper, we study a training procedure for a quadruped robot locomotion through DRL. We apply a Hierarchical Reinforcement Learning (HRL) technique based on the maximum entropy RL framework (Soft Actor-Critic, SAC) and Domain Randomization (DR) methods to train goal-reaching skills. To evaluate our method, we test the trained model on a 12-dimensional quadruped robot(D’Kitty) in simulation.