Hello everyone! 👋

We are starting a sprint which aims to expose a reasonable demo / research starting point for the quadruped platforms.

Yuxiang’s repo: CaJuN focusing on hierarchical RL / optimal control to produce agile behaviors like jumping across gaps.
Nolan’s repo: focusing on using demonstrations from motion capture including partial demos
Rosario’s repo: focusing on recovery models that bridge behavior back to the desired downstream controller/model

Since overtime we have shared a lot of code between each other, there is considerable compatibility between the three.

Our goal is to refine these code bases to provide:

  1. A reasonable, easy to deploy demo on the quadrupeds
  2. A springboard for researchers from any of the labs to use for their own research threads
  3. Transferability to any of the quadruped platforms
    1. working in the direction of generalized ground autonomy

Robots

Spot
A1 (Ghost of Lucy)
A1 ()
Go1

Table of contents