search

Healthcare-Robotics / assistive-gym

Assistive Gym, a physics-based simulation framework for physical human-robot interaction and robotic assistance.
MIT License
308 stars 75 forks source link

Action space for FeedingSawyer #21

Closed gabriansa closed 2 years ago

gabriansa commented 3 years ago

Hello,

Does anyone know what the action space for the FeedingSawyer example environment is?

Thanks in advance

Zackory commented 3 years ago

The action space is defined here: https://github.com/Healthcare-Robotics/assistive-gym/blob/main/assistive_gym/envs/env.py#L46 The actions are delta joint angles, in radians, and shown here: https://github.com/Healthcare-Robotics/assistive-gym/blob/main/assistive_gym/envs/env.py#L214 For the Sawyer, you can find the joint indices (from the URDF) that are controlled: https://github.com/Healthcare-Robotics/assistive-gym/blob/main/assistive_gym/envs/agents/sawyer.py#L8

gabriansa commented 3 years ago

Okay, I thought it could have been delta joint angles. Thanks so much for confirming.

I have another question. I am trying to train a model and then transfer the trained policy onto a real Sawyer robot. I am actually pretty close to achieving this and I guess it could be a great addition to this library. So my question is (rather dumb but just to double-check) the dimensions of the Sawyer in the simulation are realistic correct?

Just to make sure that what I input to the Sawyer is correct.

Zackory commented 3 years ago

Yes, it is the Sawyer URDF, which comes from the original Sawyer robot definition.