Humanoid Robot Gait Control Using PPO, SAC, and ES Algorithms

Gait control is the primary aspect in the case of humanoid robots as it directly has an influence on the stability and locomotion of the robot. This paper provides a comparative analysis of the three reinforcement learning algorithms namely Proximal Policy optimization (PPO), Soft-Actor Critic (SAC), and Evolutionary Strategies (ES) for the gait control in the humanoid robot. The main aim of the project is to study the effect of the behavior of the considered algorithms and make an inference on the behavior of the robot on fine-tuning the hyperparameters. The Brax environment, a simulation platform which is developed by Google and written in JAX is considered to evaluate the performance of the considered humanoid robot. Out of the considered algorithms, PPO gave the best reward of 11580.