Sim2Real Reinforcement Learning for Soccer skills
This work addresses control tasks for humanoid robots, but it is incremental as it builds on existing RL techniques and does not solve the sim2real transfer problem.
The paper tackled training humanoid robots for soccer skills using reinforcement learning, achieving more dynamic and adaptive policies for kicking, walking, and jumping that outperformed previous methods, but failed to transfer from simulation to the real world.
This thesis work presents a more efficient and effective approach to training control-related tasks for humanoid robots using Reinforcement Learning (RL). The traditional RL methods are limited in adapting to real-world environments, complexity, and natural motions, but the proposed approach overcomes these limitations by using curriculum training and Adversarial Motion Priors (AMP) technique. The results show that the developed RL policies for kicking, walking, and jumping are more dynamic, and adaptive, and outperformed previous methods. However, the transfer of the learned policy from simulation to the real world was unsuccessful, highlighting the limitations of current RL methods in fully adapting to real-world scenarios.