search

tomsilver / ploi

Code for "Planning with Learned Object Importance in Large Problem Instances using Graph Neural Networks" (AAAI 2021)
MIT License
14 stars 9 forks source link

how to run the domian "PyBullet robot domain" #1

Open hello-starry opened 2 years ago

hello-starry commented 2 years ago

Hello friends, I was reading your article recently, but I encountered some difficulties, how can I run the domain"PyBullet robot simulation" and visualize it like your paper figure 4.

Thank you very much!

tomsilver commented 2 years ago

Thanks a lot for reaching out and for your interest in PLOI!

The Pybullet environment that we used for the paper was modified from here: https://github.com/caelan/pddlstream/blob/stable/examples/pybullet/kuka/run.py Hopefully that's a useful starting point.

Unfortunately the PLOI / pybullet code was extremely hacky and messy, which is why we haven't released it yet. It was essentially hacking an entirely new pathway through the PLOI code using if statements that differentiated pddlstream vs pddlgym environments. At the end of it was a several hundred line interface file which did the environment resetting, managed timeouts, and packaged the resulting states into a pddlgym-equivalent format. We also had to change the GNN state encoding to handle continuous values.

This is a very regrettable state of affairs! I'm sorry again. It is in our long-term queue to clean and release this part of the code, but unfortunately we aren't likely to get to it anytime soon. When we do, I'll be sure to update this issue. In the meantime, let us know if we can answer any questions or be helpful in any other way.