Toni-SM

Hi @klakhi What is this intended for? CI workflows/testing?

If the purpose of the implementation is only for CI, a project can be generated using `python -c` (program passed in as string) without any modification or addition to the...

Proposal (after some discussions with @kellyguo11): Define the spaces using both: gym.spaces and python basic data types (for simplicity) indistinctly in env configs files as follows. Use the names `observation_space`,...

Hi @AndreaRossetto You can try clipping the actions explicitly or narrowing the output range (e.g.: using `tanh`), as discussed in https://github.com/isaac-sim/IsaacLab/pull/984#issuecomment-2391488483

In my opinion, RL libraries should take care of this and no Isaac Lab. Also, adding this config prevents defining other spaces since this is specific tied to `Box`. Instead,...

Hi @renezurbruegg Please, note that current implementation is in conflict with https://github.com/isaac-sim/IsaacLab/pull/1117 for the direct workflow

@renezurbruegg , as I commented previously, in my opinion the RL libraries should take care of this and no Isaac Lab. For example, using [skrl](https://skrl.readthedocs.io/) you can set model parameter...

Hi @Holiclife-KTH Currently, in the `play.py` script, the computed `actions` is the result of sampling the Gaussian distribution instead of the mean actions returned by the network (see https://skrl.readthedocs.io/en/latest/api/models/gaussian.html#concept). You...

> Is there any particular reason why the default behavior is sampled? @Mayankm96 I have not done it before because I need to add an extra step to make it...

Hi @PaoloGinefra Mmmm, could you share a link to your task or a way to repro it?