You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* NYU HPC Greene setup (#316)
* Add .env for private variables and update reqs
* Add sbatch generation script
* Add env templplate
* Combine into single sbatch script
* Add logs folde
* Version changes to make pufferlib compatible
* Add jax version
* Remove old files
* rmv old files
* Minor: Fix path
* Use typer to override default args using CLI
* Add versions
* Visualization updates and support to train on fixed dataset size (#317)
* Add support for resample_limit
* Add option to draw the log replay trajectories in visualizer
* WIP
* minimal set maps script
* minimal set maps script
* Data processing utils
* Rendering improvements and ppo
* Make dataset to resample from configurable
* Bug fix: resampling
* Agent observation figures
* Cleanup
* Prep experiments
* Prep experiments
* Make collision behavior transient (#318)
* Make collision behavior transient
* Simpler way to reset collision state for collision behavior ignore
* Reset info on collision ignore
---------
Co-authored-by: Aarav Pandya <[email protected]>
* Improved visualizer and gym environment (cloning all tensors) (#320)
* Seeds
* Improved checkpointing
* Scene datta loader
* Scene datta loader
* Change default config
* Add milliseconds in run name for launching many jobs at once
* Move exp id creating down
* Improved logging
* Add LateFusionTransformer network
* update default settings
* Draw road segments in agent obs plot
* Improve checkpointing
* Remove dead code
* Add info tensor datatype for name-based indexing
* Evaluation pipeline
* Analysis notebooks
* bug fixes
* WIP
* Separate dataset logic from environment logic
* Minor related to training
* Make sure to clone every datatype
* Eval pipeline updates and video logging
* Eval pipeline updates and video logging
* Update default constants
* Visualizer improvements
* Fix reward logging bug: Only increment rewards for living agents
* Add shuffle mode
* Make collision states for cb ignore transient
* More realistic constants
* Eval improvements
* Add videos
* Fix eval pipeline
* Minor
* Small exp changes
* Add remove non-vehicles
* Make sure data is repeated if batch_size > dataset size
* Align datasets
* Minor
* Add random baseline policy
* Fix bug in eval loop
* Small improvements
* Small vis improvements and make sure we always clone the tensors
* Reset visualizer info when we swap data batches
* wip
* Improve visualizer and rmv any bugs
* Improve visualizer and rmv any bugs
* Setting updates for verification
* Cleanup
* Remove redundant or dead code
* Init render storage
* Compare obs nb
* Ensure full deterministic data loader
* Model, various improvements and WIP
* Test notebook
* Missing function
* Add data coverage statistics
* Add data coverage statistics
* Verify that saved model makes the same mistakes
* Make sure we can resample indefenitely during training
* Data sampling bug fix: verified that current sampler is deterministic and covers the full dataset
* Add gamma as cli arg
* Evaluation pipeline improvements
* New safe branch
* Fix losses averaging bug: account for multiple epochs
* Minor
* Simplify data sampling logic.
* Average performance metrics over K done worlds
* Log truncated agents and correct bias in averaging
* Put wandb render logic in separate function
* Add option to anneal learning rate as cli arg
* Bug fix truncations
* Delete agents by agent id (#324)
* Init
* Delete agents
* Set default deleted agents idx to -1
* Reset deleted agents in setmaps
* Improve evaluation pipeline
* Support for using results df to sort scenarios for rendering
* Support for using results df to sort scenarios for rendering
* Rmv duplicates
* Important: Bootstrap rewards if truncated
* Comment out bootstrapping for A/B testing
* Minor
* Visualizer: Always plot time_step if given
* Figures for analysis
* Ap delete agents (#329)
* Init
* Delete agents
* Set default deleted agents idx to -1
* Reset deleted agents in setmaps
---------
Co-authored-by: Aarav Pandya <[email protected]>
* Comment out value function bootstrapping
* Scene manipulation experiment code
* Updated scene manipulation experiment
* Minor
* Scene Manipulation Plots (#330)
* scene manip bar plot
* selectively remove controlled/uncontrolled
* Small changes to scene manipulation experiment
* Only resample if dataset size > batch size
* Cach branch (#328)
* caching
* caching
* adding back env_torch
* adding back env_torch
* adding caching to visualiser
* adding caching and cleaning up code
* cleaning
* Revert "Cach branch (#328)"
This reverts commit 5eabb71.
* WIP
* Merge in mapname tensor
* Fix scenario filenames
* Revert filename changes
* visualize rollout trajectories
* Merge map name PR to dc/selfplay (#334)
* Add plot expert trajectories (#325)
* Add: plot expert trajectories
* Refactor: rename draw_only_controllable_veh variable
* Export map names
* Handle set maps case
* Add env helper func to extract filenames from worlds
---------
Co-authored-by: Kyungbeom Kim <[email protected]>
Co-authored-by: Daphne Cornelisse <[email protected]>
* WIP
* Revert "Merge map name PR to dc/selfplay (#334)" (#335)
This reverts commit 98166f7.
* Ap export map name (#337)
* Add plot expert trajectories (#325)
* Add: plot expert trajectories
* Refactor: rename draw_only_controllable_veh variable
* Export map names
* Handle set maps case
* Add env helper func to extract filenames from worlds
* Dont use sizeof() for loop indexing
---------
Co-authored-by: Kyungbeom Kim <[email protected]>
Co-authored-by: Daphne Cornelisse <[email protected]>
* WIP
* logging and eval improvements
* Minor
* No time to write good commit
* Add support for initialization modes
* sbatch
* change default color
* rollout viz
* backward goal experiment
* analysis
* Minor
* Minor
* Clean up configs
* Dev packaged (#345)
* Initial packaging
* Initial package rename to separate python and cpp
* Refactor python with new gpudrive namespace
* Working puffer baseline
* Change forgotten instances of gpudrive --> madrona_gpudrive
* Delete pygame renderer and package imports
* GPUDrive Reqs
* Good editable install... madrona doesn't have a build wheel option
* Update puffer baseline to latest package
* Restructure repo
* Clean up root
* Minor cleanup
* Cleanup readme
* Fix path to ppo defaults
* Remove layer of depth
* Fixes to run SB3 PPO
* Update reqs and dockerfile
* Working build
* Small readme updates
* Concise summary of key features
* Small fixes to ensure smooth puffer ppo run with rendering
* Linting
* Update readme
* Updated SB3 PPO script
* Complete installation instructions
* Minor
* Fix broken link
---------
Co-authored-by: Joseph Suarez <[email protected]>
* Update tutorials and clean up dead code (#349)
* Docker instructions
* Update tutorials
* Clean env_torch.py
* Minor
* Ensure videos are shown in place
* Remove dead code
* Update readme
* fix aspect ratio
* Add bibtex
* Add citation
* PPO loop performance improvements (#351)
* Improve settings
* Fix upper and lower bounds of gym spaces
* Mask before processing observations
* Simplify network
* Safe
* Tested ppo script
* Minor readme updates
* Minor fix to align data coverage logging
* Update paths
* colorbar
* 3d viz
* crosswalk fix
* Minor updates to work with new package
* Set render=3d to true by default
* Add heading angles to 3d boxes
* Integrate 3d rendering option with ppo script
* viz demo and 2d fix
* docstrings
* log replay in 3d
* Better Dataset Filtering (#339)
* Add plot expert trajectories (#325)
* Add: plot expert trajectories
* Refactor: rename draw_only_controllable_veh variable
* filter 3d scenes
* better processing script > less unsolvable agents
* updated viz notebook
* filter more 3d structures
* analyze reverse/u-turns
* post-processing
---------
Co-authored-by: Kyungbeom Kim <[email protected]>
* Minor readme edits and tutorial fixes
* Fix broken links
---------
Co-authored-by: Daphne Cornelisse <[email protected]>
Co-authored-by: Aarav Pandya <[email protected]>
Co-authored-by: Kevin <[email protected]>
Co-authored-by: charliemolony <[email protected]>
Co-authored-by: Kyungbeom Kim <[email protected]>
Co-authored-by: Joseph Suarez <[email protected]>
0 commit comments