search

NVlabs / motion-policy-networks

This repo has the expert data generation infrastructure and Pytorch implementation of MPiNets.
MIT License
119 stars 19 forks source link

Problem when execute run_inference.py with use_depth = True #32

Closed zichunxx closed 11 months ago

zichunxx commented 11 months ago

Hi!

Executing run_inference.py with use_depth = True seems to trigger the input error in this line, in which the correct input should be quaternion= instead of quat=.

However, I found that the input syntax error has been corrected in the main branch of robofin. So, why not update the dockerfile to install the main branch instead of v0.0.1?

Thanks!

fishbotics commented 11 months ago

Thanks for finding this bug! I fixed it in the v0.0.1 branch in robofin, so the Dockerfile should now install a version of that package without that issue.

You raise a good point about switching to the main branches. The reason why I pinned the version at a particular branch is that I want this library to be a reproducible copy of the work published in the Motion Policy Networks work and I don't want the performance to drift too far from the performance at the time of that publication. Robofin is a toolkit I'm actively developing and has an unstable API on the main branch. I suspect that switching to the main branch would both change the performance of this library and cause a lot of headaches as the API changes over time.

If you're interested in using the newer versions of robofin (or geometrout or atob), feel free to comment over on those repos and I can help you get set up.

zichunxx commented 11 months ago

Thanks for your active response, your considerations are indeed what I missed.