openpi
Fork of openpi with fixes
Star
Fork
Watch

Commits on master
Length HP sweep, looking at how long phases should be
committed
4 axis HP sweep for the advantage regularized approach
committed
Using advantage regularization in critic pretraining
committed
Per state matching and using the meta policy
committed
Changing the actor to not predict a whole action chunk but only each action, done in an open loop manner
committed
new hparam sweep for the simplified code
committed
new hparam sweep
committed
Cleaning up a bunch of stale ideas
committed
Naren Dasan
Update README
committed
Instructions on how to run the hyperparameter sweep
committed
Run multiple container pairs to one vla and then hyper parameter sweep
committed
Add forward dynamics model checkpoint to pretrained_checkpoints
committed
Training the actor as a residual on the vla
committed
Some gitignore fixes
committed
I think this is where it was working
committed
Start from bc and then work RL back in
committed
This new waypoint based method that uses bc for parts that the vla does work and RL just for the part that breaks
committed
Add a pretrained RLT encoder adaptive-bc-improved-qnet rlt-her-td-for-ac
committed
HER + RLT encoder
committed
Adding the RLT token training code
committed
Adding some launch detail to the readme, some fixes to the launch script
committed
Adding critic pretraining proceedure, sweep to collect partially solved tasks, run sgl on one task
committed
TD learning
committed
actor and critic trainer setup
committed
fix: Get the dockerfile working again
committed
Fixing the proxy to spin so to give the VLA server time to start up narendasan/opd
committed
Adding a proxy that will run the policy distilation and mixing
committed
Fixing a bunch of stuff to get Libero working with podman narendasan/libero_fixes
committed
Jimmy Tanner
docs: clarify Python 3.11 requirement for RLDS group (#910)
committed
wadeKeith
docs: clarify Python 3.11 requirement for RLDS group
committed
Kevin Black
Fix race condition in maybe_download (#846)
committed
Abhay Deshpande
Merge branch 'main' into fix-download
committed
Kevin Black
Allow configuring pytorch compilation mode (#905)
committed
Abhay D
Fix race condition in maybe_download
committed
Abhay D
ruff
committed
Abhay D
Allow configuring pytorch compilation mode
committed
Kevin Black
Fix hardcoded action dim in pi0 pytorch model (#855)
committed
Abhay D
Fix hardcoded action dim in pi0 pytorch model
committed
Jimmy Tanner
Implement gsutil download for GCS URLs (#901)
committed
Simit Pradhan
lint code
committed
Jimmy Tanner
Merge branch 'main' into 900-gcloud-download-defaults-to-authentication-even-for-public-assets
committed
Jimmy Tanner
Update codeowners (#902)
committed
Jimmy Tanner
Update codeowners
committed
Simit Pradhan
Fix local_path creation and update gsutil command
committed
Simit Pradhan
Fallback to default download method if gsutil is not available
committed
Simit Pradhan
Implement gsutil download for GCS URLs
committed
Jimmy Tanner
use EGL for headless GPU rendering in libero example (#837)
committed
Anil Zeybek
bring privileged param back
committed
Anil Zeybek
bring rendering-related params back
committed
Anil Zeybek
use EGL for headless GPU rendering in libero example
committed
Karl Pertsch
Multi-RLDS Datasets (Cotraining) + PolaRiS DROID jointpos policy configs (#821)
committed
Arhan Jain
fix ruff formatting
committed
Arhan Jain
update to public bucket paths
committed
Arhan Jain
address karl's comments
committed
Arhan Jain
remove specific paths, restore filter dict func
committed
Arhan Jain
rename paligemma config
committed
Arhan Jain
cotrain configs, RLDS filter autosearch
committed
Arhan Jain
remove no-action filter, fixing tests
committed
Arhan Jain
typing fix
committed
Arhan Jain
precommit fix again
committed