FileNotFoundError, Broken dataset assumption

[JaehongJaehongMin]

Hi,
Thank you for the kind explanation of how to reproduce your code.

Unfortunately, I've encountered some exceptions during the reproducing.
I think I should prerpocess the Peract dataset before the inference.
I've followed your instruction carefully, and don't have any version issue.

But the problem is 'eval.py' code in RLBench couldn't find the test directories are not included in your box folder.

To resolve this I've changed the name of the all_variations folder to variation-1, but had another issue with the pkl file.

The lenght of the pkl file is 429 just like the original Peract dataset, but the length of the l_sh_rgb_f (and so on) is 13 as you simplified.

Could you please share your dataset preprocessing code or explain how the handle this error?

[mlzxy]

I think I know the reason. I commented out the length check in RLBench to bypass this. This length check is unnecessary.

Could you do it just for the moment? I can think of a better way to handle this in the mean time. Also, you don't need to rename folders; consider changing them back

[mlzxy]

Specifically, comment out this check at https://github.com/stepjam/RLBench/blob/b80e51feb3694d9959cb8c0408cd385001b01382/rlbench/utils.py#L94.

Like what I did here https://github.com/mlzxy/RLBench.arp/blob/df4ee3e3f1ee977b7073cecae6673de351f95911/rlbench/utils.py#L94

I update the ENV.md to cover this.

[JaehongJaehongMin]

Thank you for the quick response!
I commented out the part of the leanght check.

But I don't think that the filename error have been resolved.

I've already restored the file name in original state.

[mlzxy]

Thanks for your feedback. I am looking into it.

[mlzxy]

Hi @JaehongJaehongMin , I have fixed the issue. There are two things you need to do:

1. Reinstall RLBench from https://github.com/mlzxy/RLBench.arp
2. Remove line "from mvt.utils import ForkedPdb" from point_renderer/rvt_renderer.py

I just realized the most frequently used RLBench (with rich language variants) comes from PerAct. This fork already becomes incompatible with the master branch of RLBench. So I rebase my change (that comment line) on top of their version.

The second point is not relevant to this issue, just one thing that I missed.

Sorry for the inconvenience. I have already updated the ENV.md to include this. Let me know if you have any problems.