This thesis aims to advance the design optimization of the RH5 humanoid robot to enhance performance in high-effort tasks, specifically focusing on the pull-up exercise to evaluate upper body strength and endurance. The research addresses existing limitations in upper-body research of humanoid robots and uses reinforcement learning (RL) as a method for enabling the RH5 to perform complex whole-body manipulations. By leveraging physical simulation environments such as Nvidia Isaac Gym, this study will develop an RL-based control policy to realize robot pull-up motions and validate the accuracy and efficiency of these motions through iterative design optimization.
This research aims to contribute to the field of humanoid robotics by providing insights into the highly dynamic motion generation through Reinforcement Learning, followed by the application of design optimization to propose an optimal upper-body design of the RH5 robot.