Booster T1 sim2real

We've been trying sim2real deployment for a policy trained with the default playground script on a Booster T1 robot.

Couldn't find any scripts in the repo so we tried to write our own. Comparing the playground "T1JoystickFlatTerrain" environment with Booster Gym, the main difference seems to be that Booster Gym does not use linear velocity as a part of their observation. We also found that Booster's low-level SDK does not provide linear velocity data, and only IMU data including RPY, Gyro and Accelerometer. Thus getting the linear velocity requires more state estimation setup on the deployment part. Booster Gym seems to support training a fully sim2real policy without needing linear velocity.

We were wondering if you'd be able to also open-source the Booster T1 deployment code. If that's too messy/not readily available, we'd be happy to write our own and open-source it, but look for your high-level guidance on whether we should do state estimation or let the policy handle that part. Interested to see if you guys tried Booster Gym and how it compares to playground. Since sim2real isn't the predominant focus of our project, we'd really appreciate any help/suggestions for any readily available pipelines and your experiences developing it.

Much appreciated in advance and hope y'all have a good one!