How to reproduce the experiments in the simulation environment

[yangsizhe]

Wonderful work!

I'm trying to reproduce the experiments in "From Imitation to Refinement Residual RL for Precise Visual Assembly". However, I'm finding there are so many scripts in this repo. I wonder which script(s) I should run to reproduce the experiments in the simulation environment. Is there a guide available for reproducing?

Thanks!

[ankile]

Hi, thank you so much for reaching out!

Yeah, it's on our immediate roadmap to set up the repo, readme, and data in such a way that anyone can run it easily, but have unfortunately not quite gotten there yet due to some recent time contraints. We plan to address this in the upcoming few days, and I'll update you once we have changes!

In short, the most important scripts to look out for are src/train/bc.py and src.train.residual_ppo.py as these contain the main training code. We will release the data shortly so that these scripts should be runnable out of the box without the need to collect one's own data.

Hopefully, you've found some use for the code despite not being quite ready to run out of the box yet.

Best, Lars

[yangsizhe]

Thank you for the reply!

[ankile]

Hi again, Sizhe,

I'm sorry for not getting back to you sooner—I've been snared up in some recent urgent matters. I will come back with more information this week.

Best, Lars

On Tue, Sep 3, 2024 at 12:28 AM Sizhe Yang @.***> wrote:

Thank you for the reply!

— Reply to this email directly, view it on GitHub https://github.com/ankile/robust-rearrangement/issues/26#issuecomment-2325580850, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIT3LITQL3OP4BNAU6FI5OTZUU3HNAVCNFSM6AAAAABMD2IH4WVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRVGU4DAOBVGA . You are receiving this because you commented.Message ID: @.***>

-- Med vennlig hilsen, Lars Lien Ankile

[ankile]

Hi!

It's not fully there yet, but the dataset and code for training the BC policies are there—and the rest is on the way.

Best, Lars