search

PaddlePaddle / MetaGym

Collection of Reinforcement Learning / Meta Reinforcement Learning Environments.
Apache License 2.0
276 stars 59 forks source link

How to run it on the real A1 robot? #50

Closed df700000 closed 2 years ago

df700000 commented 3 years ago

How to run it on the real A1 robot?

WorldEditors commented 2 years ago

RLSchool (or MetaGym) is a simulation environment. The key is that you can pre-train your control policy in the environment and then transfer it to the real robot (Sim2Real Transfer).

By "How to run it on the real A1 robot", I'm not sure if you want to ask how to transfer your learned policy from simulation to real robot, or does it mean something different?

YangXu-Insist commented 2 years ago

I have the same questions?

TomorrowIsAnOtherDay commented 2 years ago

Hello, thanks for your interest in our repository. I guess that you want to reproduce the paper that we released recently. There are some suggestions for this question: 1) follow the algorithm details in the paper and reproduce the algorithm in simulation at first. 2) deploy the trained controller in A1-robot following the APIs provided by the Unitree. 3) if the trained controller performs badly in A1, follow the details in sim-to-real section of the paper to address the problem.

Feel free to ask here if you have any more questions.

YangXu-Insist commented 2 years ago

Hello, thanks for your interest in our repository. I guess that you want to reproduce the paper that we released recently. There are some suggestions for this question:

  1. follow the algorithm details in the paper and reproduce the algorithm in simulation at first.
  2. deploy the trained controller in A1-robot following the APIs provided by the Unitree.
  3. if the trained controller performs badly in A1, follow the details in sim-to-real section of the paper to address the problem.

Feel free to ask here if you have any more questions.

OK ! Thank you very much !!!