Humanoid robots use their hyper-dexterous whole-body movements to take on tasks that require a large space, such as picking an object off the ground. This remains a challenge to to real humanoids’ high degrees of freedom. Adaptive Motion Optimization is a framework that integrates sim-to-real reinforcement learning (RL) with trajectory optimization for real-time adaptive whole-body control.
This framework was validated on a 29DOF Unitree G1 humanoid robot.
[HT] [credit:Li, Jialong and Cheng, Xuxin and Huang, Tianshu and Yang, Shiqi and Qiu, Rizhao and Wang, Xiaolong]