Training Isaac-Lift-Cube-Franka-v0 example environment doesn't succeed

[itsmehere]

I currently have the Isaac-Reach-Franka-v0 task working with the following commands:
Training: ./isaaclab.sh -p source/standalone/workflows/skrl/train.py --task Isaac-Reach-Franka-v0 --headless
Playing: ./isaaclab.sh -p source/standalone/workflows/skrl/play.py --task Isaac-Reach-Franka-v0 --num_envs=1 --checkpoint="logs/skrl/reach_franka/2025-01-18_01-23-54_ppo_torch/checkpoints/best_agent.pt"

Now, I’m trying to train the Isaac-Lift-Cube-Franka-v0 task by running very similar commands but for the franka lift environment:
Training: ./isaaclab.sh -p source/standalone/workflows/skrl/train.py --task Isaac-Lift-Cube-Franka-v0 --headless
Playing: ./isaaclab.sh -p source/standalone/workflows/skrl/play.py --task Isaac-Lift-Cube-Franka-v0 --num_envs=1 --checkpoint="logs/skrl/franka_lift/2025-01-18_02-08-25_ppo_torch/checkpoints/best_agent.pt"

The franka arm moves but doesn’t go anywhere near the cube. Am I doing something incorrect when training this particular environment, or is there some value I have to set before the lift task works? I've tried the skrl and rsl_rl libraries, but none of them seem to perform correctly. All I’ve done is setup IsaacLab and IsaacSim by following the installation instructions, and I haven’t changed any code. I’m running a few examples right off the bat, and I’m just trying to get Isaac-Lift-Cube-Franka-v0 as listed here to work. The trained policy when played does the following:

Untitled.mov

[RandomOakForest]

Thank you for posting this. I will move this to our Discussions section for the team to follow up.

[RandomOakForest]

There are other RL libraries you may try, listed in this doc. Have you tried training with RSL RL?

[itsmehere]

I have tried with both RSL RL and skrl, and they both lead to something similar :/

[RandomOakForest]

Ok, thanks, we'll take a further look.

[itsmehere]

Fixed with latest isaac sim/lab update!

[Kt981]

Which of the examples of Franka's robot enviroments have you tried and were successful?

[hejia-zhang]

Have you tried the stack env? I'm getting this error "ValueError: Could not find configuration for the environment: 'Isaac-Stack-Cube-Franka-v0'. Please check that the gym registry has the entry point: 'rl_games_cfg_entry_point'."

[hejia-zhang]

It seems it's only for imitation learning like robomimic...

[itsmehere]

I don't think the Isaac Lab repo comes with a rl_games config for the stack cubes franka task. If you want that, you'd have to add the config just like it exists for the lift task, for example, here. I'm not sure if it's that's straightforward, but that's the reason the error occurs at least. For stacking, you're right - I don't see any rl library configs in the Isaac Lab stack task by default.

[Ozzey]

You will have to add the RL config (Rl_games, SKRL or SB3) as well as a reward function. Not very straight forward since it involves writing a reward function from scratch.