StanfordVL / iGibson

A Simulation Environment to train Robots in Large Realistic Interactive Scenes
http://svl.stanford.edu/igibson
MIT License
661 stars 159 forks source link

Current state of motion primitives #188

Closed ju-kreber closed 2 years ago

ju-kreber commented 2 years ago

Hello, I'm trying to reconstruct results of the BEHAVIOR paper, in particular with reinforcement learning agents using discrete actions / motion primitives. I've come to notice that the iGibon implementation of BehaviorMPEnv (igibson.envs.behavior_mp_env) , which is used in the replay_demo_with_action_primitives_example.py in the current version of Behavior, is no longer existent. The file has been removed in 4235eef. I'm now wondering how to use motion primitives (without motion planning) in the current version. Is this no longer supported? Even if it's not supported anymore, could you hint me at a commit I could use where the functionality is still there? I've already tried multiple commits, just before the file was removed, version 2.0.5 and version 2.0.3, but none of them ran successfully due to different errors. Sadly, I could not get it working.

I'd really appreciate your advise on this, thanks!

cgokmen commented 2 years ago

Hello,

Thanks for reaching out - as you noticed we removed this motion primitive interface and are working on a new, cleaner and more effective one. Unfortunately we have no immediate date estimate for when this will be released although I would expect it to be done in May. The original interface used for those experiments was discontinued due to its incompatibility with many later refactors. I recommend trying out the earliest 2.0 releases (and will check again for you when I am at a computer later) but again this interface is not officially supported in any public version of iGibson.

Thanks, Cem

ju-kreber commented 2 years ago

Thank you for the information! I'm looking forward to the new implementation. :)

shirley-wu commented 1 year ago

Hi, may I ask how is the current state for implementing BehaviorMPEnv with primitive actions? Thanks!