Swapping the arm in an RL task

[fl1ps1de5]

I currently have written an RL algorithm that is effectively training a policy to complete the Isaac-Reach-Franka-v0 task/environment (for example).

I want to now switch to using a Kinova arm for the same task. I was wondering the easiest way to go about doing this? Is it as simple as changing the USD that is loaded, or is there a lot of custom code to be written for the task itself?

I have spent much time working on the algorithmic side of things, and less so working with IsaacLab in this capacity, so I would be very grateful for any guidance or advice.

Thanks.

[fl1ps1de5]

I suppose what I am looking for is something similar to this tutorial (modifying a direct RL environment) but for a manager based RL environment instead

[zinzuvadiadev]

Did this work?

[RandomOakForest]

Thanks for starting this discussion. That is correct, the tutorial on Modifying an existing Direct RL environment is the one to follow. Let us know if you need further help.

[fl1ps1de5]

Hey, I realise I accidentally inferred I wanted to switch from the Franka arm to the Franka arm… whereas I meant switch to the Kinova arm.

Anyways, I managed to do this by referencing the Kinova arm configuration file and within reach_env_cfg (or adjacent to it I suppose), mimicking the Franka arms joint_vel_pos.py file for the Kinova arm.

Hope that makes sense! It required a bit of manual toggling to get the reset position and scales better suited to the Kinova arm but currently seems fine.

About to do the same thing for the lifting task which should be a bit more complex, given the difference between the Jaco and Franka’s grippers.