Commit Graph

55 Commits

Author SHA1 Message Date
Alexander Soare
86365adf9f revision 2024-04-08 14:44:46 +01:00
Alexander Soare
e982c732f1 Merge remote-tracking branch 'Cadene/user/rcadene/2024_03_31_remove_torchrl' into refactor_act_remove_torchrl 2024-04-08 09:25:45 +01:00
Alexander Soare
8d2463f45b backup wip 2024-04-05 18:46:30 +01:00
Alexander Soare
1e71196fe3 backup wip 2024-04-05 17:38:29 +01:00
Alexander Soare
4863e54ce9 Merge remote-tracking branch 'Cadene/user/rcadene/2024_03_31_remove_torchrl' into refactor_act_remove_torchrl 2024-04-05 12:00:31 +01:00
Cadene
ad3379a73a fix memory leak due to itertools.cycle 2024-04-05 10:59:32 +00:00
Alexander Soare
9d77f5773d Merge remote-tracking branch 'Cadene/user/rcadene/2024_03_31_remove_torchrl' into refactor_act_remove_torchrl 2024-04-05 11:41:11 +01:00
Cadene
5af00d0c1e fix train.py, stats, eval.py (training is running) 2024-04-05 09:31:39 +00:00
Cadene
1cdfbc8b52 WIP
WIP

WIP train.py works, loss going down

WIP eval.py

Fix

WIP (eval running, TODO: verify results reproduced)

Eval works! (testing reproducibility)

WIP

pretrained model pusht reproduces same results as torchrl

pretrained model pusht reproduces same results as torchrl

Remove AbstractPolicy, Move all queues in select_action

WIP test_datasets passed (TODO: re-enable NormalizeTransform)
2024-04-04 15:31:03 +00:00
Alexander Soare
2b928eedd4 backup wip 2024-04-02 19:11:53 +01:00
Simon Alibert
c5635b7d94 Minor fixes for #47 2024-03-25 18:50:47 +01:00
Simon Alibert
7cdd6d2450 Renamed set_seed -> set_global_seed 2024-03-25 17:19:28 +01:00
Cadene
be6364f109 fix, it's training now! 2024-03-25 12:28:07 +01:00
Alexander Soare
bd40ffc53c revision 2024-03-22 15:43:45 +00:00
Alexander Soare
15ff3b3af8 add fixes for reproducibility 2024-03-22 15:06:57 +00:00
Alexander Soare
72d3c3120b Merge remote-tracking branch 'upstream/main' into fix_pusht_diffusion 2024-03-21 10:20:52 +00:00
Simon Alibert
4631d36c05 Add get_safe_torch_device in policies 2024-03-20 18:38:55 +01:00
Alexander Soare
d323993569 backup wip 2024-03-20 15:01:27 +00:00
Alexander Soare
4b7ec81dde remove abstracmethods, fix online training 2024-03-20 14:49:41 +00:00
Alexander Soare
ea17f4ce50 backup wip 2024-03-19 16:02:09 +00:00
Alexander Soare
09ddd9bf92 Merge branch 'main' into user/alexander-soare/multistep_policy_and_serial_env 2024-03-18 18:27:50 +00:00
Alexander Soare
8e346b379d switch between train and eval 2024-03-18 09:45:17 +00:00
Alexander Soare
ba91976944 wip: still needs batch logic for act and tdmp 2024-03-14 15:24:10 +00:00
Remi Cadene
9d002032d1 Add Aloha env and ACT policy
WIP Aloha env tests pass

Rendering works (fps look fast tho? TODO action bounding is too wide [-1,1])

Update README

Copy past from act repo

Remove download.py add a WIP for Simxarm

Remove download.py add a WIP for Simxarm

Add act yaml (TODO: try train.py)

Training can runs (TODO: eval)

Add tasks without end_effector that are compatible with dataset, Eval can run (TODO: training and pretrained model)

Add AbstractEnv, Refactor AlohaEnv, Add rendering_hook in env, Minor modifications, (TODO: Refactor Pusht and Simxarm)

poetry lock

fix bug in compute_stats for action normalization

fix more bugs in normalization

fix training

fix import

PushtEnv inheriates AbstractEnv, Improve factory Normalization

Add _make_env to EnvAbstract

Add call_rendering_hooks to pusht env

SimxarmEnv inherites from AbstractEnv (NOT TESTED)

Add aloha tests artifacts + update pusht stats

fix image normalization: before env was in [0,1] but dataset in [0,255], and now both in [0,255]

Small fix on simxarm

Add next to obs

Add top camera to Aloha env (TODO: make it compatible with set of cameras)

Add top camera to Aloha env (TODO: make it compatible with set of cameras)
2024-03-12 10:27:48 +00:00
Cadene
816b2e9d63 fix more bugs in normalization 2024-03-11 11:03:51 +00:00
Remi Cadene
f95ecd66fc Improve visualize_dataset, Improve AbstractReplayBuffer, Small improvements 2024-03-06 10:15:57 +00:00
Remi Cadene
2bcf2631b9 minor comment 2024-03-04 22:34:44 +00:00
Remi
e990f3e148 Merge pull request #6 from Cadene/user/rcadene/2024_03_04_diffusion
Make diffusion work
2024-03-04 18:30:40 +01:00
Remi Cadene
cfc304e870 Refactor env queue, Training diffusion works (Still not converging) 2024-03-04 11:00:51 +00:00
Remi Cadene
4c400b41a5 Improve log msg in train.py 2024-03-03 13:22:09 +00:00
Simon Alibert
b859e89936 Fix for PR #5 2024-03-03 13:05:21 +01:00
Simon Alibert
b33ec5a630 Add run on cpu-only compatibility 2024-03-03 12:47:26 +01:00
Remi Cadene
80785f8d0e Small fix, Refactor diffusion, Diffusion runs (TODO: remove normalization in diffusion) 2024-03-02 17:04:39 +00:00
Remi Cadene
1ae6205269 Add Normalize, non_blocking=True in tdmpc, tdmpc run (TODO: diffusion) 2024-03-02 15:53:29 +00:00
Cadene
ae050d2e94 Solve conflicts + pre-commit run -a 2024-02-29 23:31:32 +00:00
Cadene
0b9027f05e Clean logging, Refactor 2024-02-29 23:21:27 +00:00
Simon Alibert
7e024fdce6 Ran pre-commit run --all-files 2024-02-29 13:37:48 +01:00
Cadene
cf5063e50e Add diffusion policy (train and eval works, TODO: reproduce results) 2024-02-28 15:21:42 +00:00
Cadene
e543c9a42c small fix % 2024-02-27 11:54:31 +00:00
Cadene
7df542445c Small fix and improve logging message 2024-02-27 11:44:26 +00:00
Cadene
21670dce90 Refactor train, eval_policy, logger, Add diffusion.yaml (WIP) 2024-02-26 01:10:09 +00:00
Cadene
b16c334825 Refactor configs to have env in seperate yaml + Fix training 2024-02-25 17:42:47 +00:00
Cadene
ed80db2846 Sanitize cfg.env 2024-02-25 12:02:29 +00:00
Cadene
598bb496b0 Add policies/factory, Add test, Add _self_ in config 2024-02-25 10:50:23 +00:00
Cadene
aed02dc7c6 Add multithreading for video generation, Speed policy sampling 2024-02-24 18:18:39 +00:00
Cadene
63d18475cc fix simxarm factory 2024-02-22 13:04:24 +00:00
Cadene
e3643d6146 Wandb works, One output dir 2024-02-22 12:14:12 +00:00
Cadene
3dc14b5576 Add Prod transform, Add test_factory 2024-02-20 14:22:16 +00:00
Cadene
3da6ffb2cb Fix unit tests, Refactor, Add pusht env, (TODO pusht replay buffer, image preprocessing) 2024-02-20 12:26:57 +00:00
Cadene
a5c305a7a4 offline training + online finetuning converge to 33 reward! 2024-02-18 01:23:44 +00:00