Locomotion

There’s a lot of cool work in locomotion, but it seems most the field is solved. The main issue is that we know the Kinematics through the MoCAP, but we have no understanding of the Dynamics. We use RL to help us learn the dynamics.

Leaders

• Xue Bin Peng
• Tairan He
• Chong Zhang
• Guy Tevet

These idk

• BeyondMimic
• InterMimic

Here are some papers

• DeepMimic

long horizon locomotion

• Perpetual Humanoid Control for Realtime Simulated Avatars
• Universal Humanoid Motion Representations for PhysicsBased Control
• Diffuse-CLoC

Whole body humanoid stuff?:

• CLONE
• TWIST
• ASAP

Motion imitation (non humanoid) stuff

• AMP Adversarial Motion Priors for Stylized PhysicsBased Character Control
• ASE
• MotionVAE
• Flexible Motion Inbetweening with Diffusion Models
• CLoSD Closing the Loop between Simulation and Diffusion for multitask character control
• Human Motion Diffusion Model

What is the general flow of locomotion policies?

• Your policy outputs joint positions, and you have a low-level PD controller that translates those joint position targets into actual motor torques to be sent out