Bachelor’s Graduation Project

《Complex Motion Skill Learning and Generalization of Humanoid Robots Based on Imitation Learning and Reinforcement Learning》(Based on ASAP).

Database: AMASS

After completing the SMPL shape and motion preparation, as well as the Humanoid-SMPL fitting and retargeting, I proceeded to train the policies. The results are shown below. The reference humanoid motions are displayed in MuJoCo, while the training results are simulated in Isaac Gym. Each policy was trained for approximately 8800 to 17000 iterations, taking about 22–40 hours on an RTX 3090.