solarian-diffusion 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- diffusion_policy/__init__.py +0 -0
- diffusion_policy/codecs/__init__.py +0 -0
- diffusion_policy/codecs/imagecodecs_numcodecs.py +1408 -0
- diffusion_policy/common/__init__.py +0 -0
- diffusion_policy/common/checkpoint_util.py +64 -0
- diffusion_policy/common/json_logger.py +117 -0
- diffusion_policy/common/normalize_util.py +251 -0
- diffusion_policy/common/pose_repr_util.py +112 -0
- diffusion_policy/common/pytorch_util.py +82 -0
- diffusion_policy/common/q_critic_util.py +135 -0
- diffusion_policy/common/replay_buffer.py +672 -0
- diffusion_policy/common/sampler.py +204 -0
- diffusion_policy/config/legacy/task/blockpush_lowdim_seed.yaml +34 -0
- diffusion_policy/config/legacy/task/blockpush_lowdim_seed_abs.yaml +34 -0
- diffusion_policy/config/legacy/task/can_image.yaml +64 -0
- diffusion_policy/config/legacy/task/can_image_abs.yaml +64 -0
- diffusion_policy/config/legacy/task/can_lowdim.yaml +45 -0
- diffusion_policy/config/legacy/task/can_lowdim_abs.yaml +46 -0
- diffusion_policy/config/legacy/task/kitchen_lowdim.yaml +33 -0
- diffusion_policy/config/legacy/task/kitchen_lowdim_abs.yaml +38 -0
- diffusion_policy/config/legacy/task/lift_abs.yaml +72 -0
- diffusion_policy/config/legacy/task/lift_image.yaml +64 -0
- diffusion_policy/config/legacy/task/lift_image_abs.yaml +63 -0
- diffusion_policy/config/legacy/task/lift_lowdim.yaml +46 -0
- diffusion_policy/config/legacy/task/lift_lowdim_abs.yaml +47 -0
- diffusion_policy/config/legacy/task/pusht_image.yaml +40 -0
- diffusion_policy/config/legacy/task/pusht_lowdim.yaml +34 -0
- diffusion_policy/config/legacy/task/real_pusht_image.yaml +47 -0
- diffusion_policy/config/legacy/task/square_abs.yaml +72 -0
- diffusion_policy/config/legacy/task/square_image.yaml +64 -0
- diffusion_policy/config/legacy/task/square_image_abs.yaml +64 -0
- diffusion_policy/config/legacy/task/square_lowdim.yaml +46 -0
- diffusion_policy/config/legacy/task/square_lowdim_abs.yaml +47 -0
- diffusion_policy/config/legacy/task/tool_hang_abs.yaml +71 -0
- diffusion_policy/config/legacy/task/tool_hang_image.yaml +63 -0
- diffusion_policy/config/legacy/task/tool_hang_image_abs.yaml +63 -0
- diffusion_policy/config/legacy/task/tool_hang_lowdim.yaml +45 -0
- diffusion_policy/config/legacy/task/tool_hang_lowdim_abs.yaml +46 -0
- diffusion_policy/config/legacy/task/transport_image.yaml +75 -0
- diffusion_policy/config/legacy/task/transport_image_abs.yaml +75 -0
- diffusion_policy/config/legacy/task/transport_lowdim.yaml +49 -0
- diffusion_policy/config/legacy/task/transport_lowdim_abs.yaml +49 -0
- diffusion_policy/config/legacy/task/umi.yaml +45 -0
- diffusion_policy/config/legacy/task/umi_image.yaml +36 -0
- diffusion_policy/config/legacy/task/umi_teleop_image.yaml +41 -0
- diffusion_policy/config/legacy/train_bet_lowdim_workspace.yaml +132 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_hybrid_workspace.yaml +141 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_lowdim_kitchen_workspace.yaml +146 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_lowdim_pusht_workspace.yaml +146 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_lowdim_workspace.yaml +146 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_real_hybrid_workspace.yaml +144 -0
- diffusion_policy/config/legacy/train_diffusion_transformer_umi_teleop_hybrid_workspace.yaml +156 -0
- diffusion_policy/config/legacy/train_diffusion_unet_ddim_hybrid_workspace.yaml +140 -0
- diffusion_policy/config/legacy/train_diffusion_unet_ddim_lowdim_workspace.yaml +146 -0
- diffusion_policy/config/legacy/train_diffusion_unet_hybrid_workspace.yaml +137 -0
- diffusion_policy/config/legacy/train_diffusion_unet_image_pretrained_workspace.yaml +149 -0
- diffusion_policy/config/legacy/train_diffusion_unet_image_workspace.yaml +150 -0
- diffusion_policy/config/legacy/train_diffusion_unet_lowdim_workspace.yaml +143 -0
- diffusion_policy/config/legacy/train_diffusion_unet_real_hybrid_workspace.yaml +141 -0
- diffusion_policy/config/legacy/train_diffusion_unet_real_image_workspace.yaml +152 -0
- diffusion_policy/config/legacy/train_diffusion_unet_real_pretrained_workspace.yaml +152 -0
- diffusion_policy/config/legacy/train_diffusion_unet_timm_workspace.yaml +145 -0
- diffusion_policy/config/legacy/train_diffusion_unet_umi_hybrid_workspace.yaml +141 -0
- diffusion_policy/config/legacy/train_diffusion_unet_umi_teleop_hybrid_workspace.yaml +153 -0
- diffusion_policy/config/legacy/train_diffusion_unet_umi_timm_workspace.yaml +145 -0
- diffusion_policy/config/legacy/train_diffusion_unet_umi_workspace.yaml +150 -0
- diffusion_policy/config/legacy/train_diffusion_unet_video_workspace.yaml +128 -0
- diffusion_policy/config/legacy/train_ibc_dfo_hybrid_workspace.yaml +111 -0
- diffusion_policy/config/legacy/train_ibc_dfo_lowdim_workspace.yaml +109 -0
- diffusion_policy/config/legacy/train_ibc_dfo_real_hybrid_workspace.yaml +111 -0
- diffusion_policy/config/legacy/train_robomimic_image_workspace.yaml +92 -0
- diffusion_policy/config/legacy/train_robomimic_lowdim_workspace.yaml +92 -0
- diffusion_policy/config/legacy/train_robomimic_real_image_workspace.yaml +92 -0
- diffusion_policy/config/rl_annotate.yaml +265 -0
- diffusion_policy/config/rl_annotate_per_frame.yaml +162 -0
- diffusion_policy/config/rl_annotate_per_frame_v4.yaml +162 -0
- diffusion_policy/config/rl_annotate_v2.yaml +141 -0
- diffusion_policy/config/rl_annotate_v3.yaml +165 -0
- diffusion_policy/config/task/lift.yaml +89 -0
- diffusion_policy/config/task/square.yaml +92 -0
- diffusion_policy/config/task/tool_hang.yaml +93 -0
- diffusion_policy/config/task/tool_hang_abs.yaml +71 -0
- diffusion_policy/config/task/umi.yaml +105 -0
- diffusion_policy/config/task/umi_bimanual.yaml +135 -0
- diffusion_policy/config/task/umi_cup_wild.yaml +106 -0
- diffusion_policy/config/task/umi_image.yaml +36 -0
- diffusion_policy/config/task/umi_pick_lego.yaml +106 -0
- diffusion_policy/config/task/umi_pick_single_lego_175.yaml +106 -0
- diffusion_policy/config/task/umi_pick_single_lego_bbox.yaml +135 -0
- diffusion_policy/config/task/umi_teleop_image.yaml +41 -0
- diffusion_policy/config/train_diffusion_transformer_umi_bimanual_workspace.yaml +185 -0
- diffusion_policy/config/train_diffusion_transformer_umi_workspace.yaml +185 -0
- diffusion_policy/config/train_diffusion_unet_image_workspace.yaml +152 -0
- diffusion_policy/config/train_diffusion_unet_timm_robomimic_workspace.yaml +152 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_cup_wild.yaml +260 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_lego_pile_bbox.yaml +265 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_lego_pile_bbox_awr.yaml +324 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_lego_pile_bbox_parl.yaml +399 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_single_lego.yaml +258 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_single_lego_125.yaml +260 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_single_lego_175.yaml +260 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_single_lego_bbox.yaml +265 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_pick_two_legos_bbox.yaml +265 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_workspace.yaml +224 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_workspace_lego.yaml +213 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_workspace_lego_resume.yaml +213 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_workspace_pick_lego.yaml +254 -0
- diffusion_policy/config/train_diffusion_unet_timm_umi_workspace_red_cup.yaml +219 -0
- diffusion_policy/config/train_diffusion_unet_umi_bimanual_workspace.yaml +203 -0
- diffusion_policy/config/train_iql_umi.yaml +186 -0
- diffusion_policy/config/train_iql_umi_v2.yaml +212 -0
- diffusion_policy/config/train_iql_umi_v3.yaml +212 -0
- diffusion_policy/dataset/__init__.py +0 -0
- diffusion_policy/dataset/base_dataset.py +76 -0
- diffusion_policy/dataset/umi_dataset.py +652 -0
- diffusion_policy/env_runner/__init__.py +0 -0
- diffusion_policy/env_runner/base_image_runner.py +9 -0
- diffusion_policy/env_runner/real_pusht_image_runner.py +10 -0
- diffusion_policy/model/__init__.py +0 -0
- diffusion_policy/model/common/__init__.py +0 -0
- diffusion_policy/model/common/dict_of_tensor_mixin.py +38 -0
- diffusion_policy/model/common/lr_scheduler.py +46 -0
- diffusion_policy/model/common/module_attr_mixin.py +14 -0
- diffusion_policy/model/common/normalizer.py +353 -0
- diffusion_policy/model/common/tensor_util.py +960 -0
- diffusion_policy/model/diffusion/__init__.py +0 -0
- diffusion_policy/model/diffusion/conditional_unet1d.py +238 -0
- diffusion_policy/model/diffusion/conv1d_components.py +46 -0
- diffusion_policy/model/diffusion/ema_model.py +88 -0
- diffusion_policy/model/diffusion/mask_generator.py +221 -0
- diffusion_policy/model/diffusion/positional_embedding.py +17 -0
- diffusion_policy/model/vision/__init__.py +0 -0
- diffusion_policy/model/vision/crop_randomizer.py +288 -0
- diffusion_policy/model/vision/multi_image_obs_encoder.py +195 -0
- diffusion_policy/model/vision/timm_obs_encoder.py +525 -0
- diffusion_policy/policy/__init__.py +0 -0
- diffusion_policy/policy/base_image_policy.py +27 -0
- diffusion_policy/policy/diffusion_unet_image_policy.py +264 -0
- diffusion_policy/policy/diffusion_unet_timm_policy.py +607 -0
- diffusion_policy/policy/parl_action_optimizer.py +675 -0
- diffusion_policy/workspace/__init__.py +0 -0
- diffusion_policy/workspace/base_workspace.py +146 -0
- diffusion_policy/workspace/train_diffusion_unet_image_workspace.py +1585 -0
- solarian_diffusion/__init__.py +32 -0
- solarian_diffusion/config.py +31 -0
- solarian_diffusion/policy_server.py +316 -0
- solarian_diffusion/rtc.py +272 -0
- solarian_diffusion/service.py +16 -0
- solarian_diffusion/trainer.py +213 -0
- solarian_diffusion-0.1.0.dist-info/METADATA +126 -0
- solarian_diffusion-0.1.0.dist-info/RECORD +159 -0
- solarian_diffusion-0.1.0.dist-info/WHEEL +5 -0
- solarian_diffusion-0.1.0.dist-info/entry_points.txt +8 -0
- solarian_diffusion-0.1.0.dist-info/licenses/LICENSE +201 -0
- solarian_diffusion-0.1.0.dist-info/licenses/NOTICE +45 -0
- solarian_diffusion-0.1.0.dist-info/top_level.txt +3 -0
- umi/__init__.py +0 -0
- umi/common/__init__.py +0 -0
- umi/common/pose_util.py +115 -0
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
name: lift_image
|
|
2
|
+
|
|
3
|
+
shape_meta: &shape_meta
|
|
4
|
+
# acceptable types: rgb, low_dim
|
|
5
|
+
obs:
|
|
6
|
+
agentview_image:
|
|
7
|
+
shape: [3, 84, 84]
|
|
8
|
+
type: rgb
|
|
9
|
+
robot0_eye_in_hand_image:
|
|
10
|
+
shape: [3, 84, 84]
|
|
11
|
+
type: rgb
|
|
12
|
+
robot0_eef_pos:
|
|
13
|
+
shape: [3]
|
|
14
|
+
# type default: low_dim
|
|
15
|
+
robot0_eef_quat:
|
|
16
|
+
shape: [4]
|
|
17
|
+
robot0_gripper_qpos:
|
|
18
|
+
shape: [2]
|
|
19
|
+
action:
|
|
20
|
+
shape: [7]
|
|
21
|
+
|
|
22
|
+
task_name: &task_name lift
|
|
23
|
+
dataset_type: &dataset_type ph
|
|
24
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/image.hdf5
|
|
25
|
+
abs_action: &abs_action False
|
|
26
|
+
|
|
27
|
+
env_runner:
|
|
28
|
+
_target_: diffusion_policy.env_runner.robomimic_image_runner.RobomimicImageRunner
|
|
29
|
+
dataset_path: *dataset_path
|
|
30
|
+
shape_meta: *shape_meta
|
|
31
|
+
# costs 1GB per env
|
|
32
|
+
n_train: 6
|
|
33
|
+
n_train_vis: 1
|
|
34
|
+
train_start_idx: 0
|
|
35
|
+
n_test: 50
|
|
36
|
+
n_test_vis: 3
|
|
37
|
+
test_start_seed: 100000
|
|
38
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
39
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
40
|
+
n_obs_steps: ${n_obs_steps}
|
|
41
|
+
n_action_steps: ${n_action_steps}
|
|
42
|
+
render_obs_key: 'agentview_image'
|
|
43
|
+
fps: 10
|
|
44
|
+
crf: 22
|
|
45
|
+
past_action: ${past_action_visible}
|
|
46
|
+
abs_action: *abs_action
|
|
47
|
+
tqdm_interval_sec: 1.0
|
|
48
|
+
n_envs: 28
|
|
49
|
+
# evaluation at this config requires a 16 core 64GB instance.
|
|
50
|
+
|
|
51
|
+
dataset:
|
|
52
|
+
_target_: diffusion_policy.dataset.robomimic_replay_image_dataset.RobomimicReplayImageDataset
|
|
53
|
+
shape_meta: *shape_meta
|
|
54
|
+
dataset_path: *dataset_path
|
|
55
|
+
horizon: ${horizon}
|
|
56
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
57
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
58
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
59
|
+
abs_action: *abs_action
|
|
60
|
+
rotation_rep: 'rotation_6d'
|
|
61
|
+
use_legacy_normalizer: False
|
|
62
|
+
use_cache: True
|
|
63
|
+
seed: 42
|
|
64
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,63 @@
|
|
|
1
|
+
name: lift_image
|
|
2
|
+
|
|
3
|
+
shape_meta: &shape_meta
|
|
4
|
+
# acceptable types: rgb, low_dim
|
|
5
|
+
obs:
|
|
6
|
+
agentview_image:
|
|
7
|
+
shape: [3, 84, 84]
|
|
8
|
+
type: rgb
|
|
9
|
+
robot0_eye_in_hand_image:
|
|
10
|
+
shape: [3, 84, 84]
|
|
11
|
+
type: rgb
|
|
12
|
+
robot0_eef_pos:
|
|
13
|
+
shape: [3]
|
|
14
|
+
# type default: low_dim
|
|
15
|
+
robot0_eef_quat:
|
|
16
|
+
shape: [4]
|
|
17
|
+
robot0_gripper_qpos:
|
|
18
|
+
shape: [2]
|
|
19
|
+
action:
|
|
20
|
+
shape: [10]
|
|
21
|
+
|
|
22
|
+
task_name: &task_name lift
|
|
23
|
+
dataset_type: &dataset_type ph
|
|
24
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/image_abs.hdf5
|
|
25
|
+
abs_action: &abs_action True
|
|
26
|
+
|
|
27
|
+
env_runner:
|
|
28
|
+
_target_: diffusion_policy.env_runner.robomimic_image_runner.RobomimicImageRunner
|
|
29
|
+
dataset_path: *dataset_path
|
|
30
|
+
shape_meta: *shape_meta
|
|
31
|
+
n_train: 6
|
|
32
|
+
n_train_vis: 2
|
|
33
|
+
train_start_idx: 0
|
|
34
|
+
n_test: 50
|
|
35
|
+
n_test_vis: 4
|
|
36
|
+
test_start_seed: 100000
|
|
37
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
38
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
39
|
+
n_obs_steps: ${n_obs_steps}
|
|
40
|
+
n_action_steps: ${n_action_steps}
|
|
41
|
+
render_obs_key: 'agentview_image'
|
|
42
|
+
fps: 10
|
|
43
|
+
crf: 22
|
|
44
|
+
past_action: ${past_action_visible}
|
|
45
|
+
abs_action: *abs_action
|
|
46
|
+
tqdm_interval_sec: 1.0
|
|
47
|
+
n_envs: 28
|
|
48
|
+
# evaluation at this config requires a 16 core 64GB instance.
|
|
49
|
+
|
|
50
|
+
dataset:
|
|
51
|
+
_target_: diffusion_policy.dataset.robomimic_replay_image_dataset.RobomimicReplayImageDataset
|
|
52
|
+
shape_meta: *shape_meta
|
|
53
|
+
dataset_path: *dataset_path
|
|
54
|
+
horizon: ${horizon}
|
|
55
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
56
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
57
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
58
|
+
abs_action: *abs_action
|
|
59
|
+
rotation_rep: 'rotation_6d'
|
|
60
|
+
use_legacy_normalizer: False
|
|
61
|
+
use_cache: True
|
|
62
|
+
seed: 42
|
|
63
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,46 @@
|
|
|
1
|
+
name: lift_lowdim
|
|
2
|
+
|
|
3
|
+
obs_dim: 19
|
|
4
|
+
action_dim: 7
|
|
5
|
+
keypoint_dim: 3
|
|
6
|
+
|
|
7
|
+
obs_keys: &obs_keys ['object', 'robot0_eef_pos', 'robot0_eef_quat', 'robot0_gripper_qpos']
|
|
8
|
+
task_name: &task_name lift
|
|
9
|
+
dataset_type: &dataset_type ph
|
|
10
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/low_dim.hdf5
|
|
11
|
+
abs_action: &abs_action False
|
|
12
|
+
|
|
13
|
+
env_runner:
|
|
14
|
+
_target_: diffusion_policy.env_runner.robomimic_lowdim_runner.RobomimicLowdimRunner
|
|
15
|
+
dataset_path: *dataset_path
|
|
16
|
+
obs_keys: *obs_keys
|
|
17
|
+
n_train: 6
|
|
18
|
+
n_train_vis: 2
|
|
19
|
+
train_start_idx: 0
|
|
20
|
+
n_test: 50
|
|
21
|
+
n_test_vis: 4
|
|
22
|
+
test_start_seed: 100000
|
|
23
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
24
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
25
|
+
n_obs_steps: ${n_obs_steps}
|
|
26
|
+
n_action_steps: ${n_action_steps}
|
|
27
|
+
n_latency_steps: ${n_latency_steps}
|
|
28
|
+
render_hw: [128,128]
|
|
29
|
+
fps: 10
|
|
30
|
+
crf: 22
|
|
31
|
+
past_action: ${past_action_visible}
|
|
32
|
+
abs_action: *abs_action
|
|
33
|
+
tqdm_interval_sec: 1.0
|
|
34
|
+
n_envs: 28
|
|
35
|
+
|
|
36
|
+
dataset:
|
|
37
|
+
_target_: diffusion_policy.dataset.robomimic_replay_lowdim_dataset.RobomimicReplayLowdimDataset
|
|
38
|
+
dataset_path: *dataset_path
|
|
39
|
+
horizon: ${horizon}
|
|
40
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
41
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
42
|
+
obs_keys: *obs_keys
|
|
43
|
+
abs_action: *abs_action
|
|
44
|
+
use_legacy_normalizer: False
|
|
45
|
+
seed: 42
|
|
46
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
name: lift_lowdim
|
|
2
|
+
|
|
3
|
+
obs_dim: 19
|
|
4
|
+
action_dim: 10
|
|
5
|
+
keypoint_dim: 3
|
|
6
|
+
|
|
7
|
+
obs_keys: &obs_keys ['object', 'robot0_eef_pos', 'robot0_eef_quat', 'robot0_gripper_qpos']
|
|
8
|
+
task_name: &task_name lift
|
|
9
|
+
dataset_type: &dataset_type ph
|
|
10
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/low_dim_abs.hdf5
|
|
11
|
+
abs_action: &abs_action True
|
|
12
|
+
|
|
13
|
+
env_runner:
|
|
14
|
+
_target_: diffusion_policy.env_runner.robomimic_lowdim_runner.RobomimicLowdimRunner
|
|
15
|
+
dataset_path: *dataset_path
|
|
16
|
+
obs_keys: *obs_keys
|
|
17
|
+
n_train: 6
|
|
18
|
+
n_train_vis: 2
|
|
19
|
+
train_start_idx: 0
|
|
20
|
+
n_test: 50
|
|
21
|
+
n_test_vis: 3
|
|
22
|
+
test_start_seed: 100000
|
|
23
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
24
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
25
|
+
n_obs_steps: ${n_obs_steps}
|
|
26
|
+
n_action_steps: ${n_action_steps}
|
|
27
|
+
n_latency_steps: ${n_latency_steps}
|
|
28
|
+
render_hw: [128,128]
|
|
29
|
+
fps: 10
|
|
30
|
+
crf: 22
|
|
31
|
+
past_action: ${past_action_visible}
|
|
32
|
+
abs_action: *abs_action
|
|
33
|
+
tqdm_interval_sec: 1.0
|
|
34
|
+
n_envs: 28
|
|
35
|
+
|
|
36
|
+
dataset:
|
|
37
|
+
_target_: diffusion_policy.dataset.robomimic_replay_lowdim_dataset.RobomimicReplayLowdimDataset
|
|
38
|
+
dataset_path: *dataset_path
|
|
39
|
+
horizon: ${horizon}
|
|
40
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
41
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
42
|
+
obs_keys: *obs_keys
|
|
43
|
+
abs_action: *abs_action
|
|
44
|
+
use_legacy_normalizer: False
|
|
45
|
+
rotation_rep: rotation_6d
|
|
46
|
+
seed: 42
|
|
47
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
name: pusht_image
|
|
2
|
+
|
|
3
|
+
image_shape: &image_shape [3, 96, 96]
|
|
4
|
+
shape_meta: &shape_meta
|
|
5
|
+
# acceptable types: rgb, low_dim
|
|
6
|
+
obs:
|
|
7
|
+
image:
|
|
8
|
+
shape: *image_shape
|
|
9
|
+
type: rgb
|
|
10
|
+
agent_pos:
|
|
11
|
+
shape: [2]
|
|
12
|
+
type: low_dim
|
|
13
|
+
action:
|
|
14
|
+
shape: [2]
|
|
15
|
+
|
|
16
|
+
env_runner:
|
|
17
|
+
_target_: diffusion_policy.env_runner.pusht_image_runner.PushTImageRunner
|
|
18
|
+
n_train: 6
|
|
19
|
+
n_train_vis: 2
|
|
20
|
+
train_start_seed: 0
|
|
21
|
+
n_test: 50
|
|
22
|
+
n_test_vis: 4
|
|
23
|
+
legacy_test: True
|
|
24
|
+
test_start_seed: 100000
|
|
25
|
+
max_steps: 300
|
|
26
|
+
n_obs_steps: ${n_obs_steps}
|
|
27
|
+
n_action_steps: ${n_action_steps}
|
|
28
|
+
fps: 10
|
|
29
|
+
past_action: ${past_action_visible}
|
|
30
|
+
n_envs: null
|
|
31
|
+
|
|
32
|
+
dataset:
|
|
33
|
+
_target_: diffusion_policy.dataset.pusht_image_dataset.PushTImageDataset
|
|
34
|
+
zarr_path: data/pusht/pusht_cchi_v7_replay.zarr
|
|
35
|
+
horizon: ${horizon}
|
|
36
|
+
pad_before: ${eval:'${n_obs_steps}-1'}
|
|
37
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
38
|
+
seed: 42
|
|
39
|
+
val_ratio: 0.02
|
|
40
|
+
max_train_episodes: 90
|
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
name: pusht_lowdim
|
|
2
|
+
|
|
3
|
+
obs_dim: 20 # 9*2 keypoints + 2 state
|
|
4
|
+
action_dim: 2
|
|
5
|
+
keypoint_dim: 2
|
|
6
|
+
|
|
7
|
+
env_runner:
|
|
8
|
+
_target_: diffusion_policy.env_runner.pusht_keypoints_runner.PushTKeypointsRunner
|
|
9
|
+
keypoint_visible_rate: ${keypoint_visible_rate}
|
|
10
|
+
n_train: 6
|
|
11
|
+
n_train_vis: 2
|
|
12
|
+
train_start_seed: 0
|
|
13
|
+
n_test: 50
|
|
14
|
+
n_test_vis: 4
|
|
15
|
+
legacy_test: True
|
|
16
|
+
test_start_seed: 100000
|
|
17
|
+
max_steps: 300
|
|
18
|
+
n_obs_steps: ${n_obs_steps}
|
|
19
|
+
n_action_steps: ${n_action_steps}
|
|
20
|
+
n_latency_steps: ${n_latency_steps}
|
|
21
|
+
fps: 10
|
|
22
|
+
agent_keypoints: False
|
|
23
|
+
past_action: ${past_action_visible}
|
|
24
|
+
n_envs: null
|
|
25
|
+
|
|
26
|
+
dataset:
|
|
27
|
+
_target_: diffusion_policy.dataset.pusht_dataset.PushTLowdimDataset
|
|
28
|
+
zarr_path: data/pusht/pusht_cchi_v7_replay.zarr
|
|
29
|
+
horizon: ${horizon}
|
|
30
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
31
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
32
|
+
seed: 42
|
|
33
|
+
val_ratio: 0.02
|
|
34
|
+
max_train_episodes: 90
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
name: real_image
|
|
2
|
+
|
|
3
|
+
image_shape: [3, 240, 320]
|
|
4
|
+
dataset_path: data/pusht_real/real_pusht_20230105
|
|
5
|
+
|
|
6
|
+
shape_meta: &shape_meta
|
|
7
|
+
# acceptable types: rgb, low_dim
|
|
8
|
+
obs:
|
|
9
|
+
# camera_0:
|
|
10
|
+
# shape: ${task.image_shape}
|
|
11
|
+
# type: rgb
|
|
12
|
+
camera_1:
|
|
13
|
+
shape: ${task.image_shape}
|
|
14
|
+
type: rgb
|
|
15
|
+
# camera_2:
|
|
16
|
+
# shape: ${task.image_shape}
|
|
17
|
+
# type: rgb
|
|
18
|
+
camera_3:
|
|
19
|
+
shape: ${task.image_shape}
|
|
20
|
+
type: rgb
|
|
21
|
+
# camera_4:
|
|
22
|
+
# shape: ${task.image_shape}
|
|
23
|
+
# type: rgb
|
|
24
|
+
robot_eef_pose:
|
|
25
|
+
shape: [2]
|
|
26
|
+
type: low_dim
|
|
27
|
+
action:
|
|
28
|
+
shape: [2]
|
|
29
|
+
|
|
30
|
+
env_runner:
|
|
31
|
+
_target_: diffusion_policy.env_runner.real_pusht_image_runner.RealPushTImageRunner
|
|
32
|
+
|
|
33
|
+
dataset:
|
|
34
|
+
_target_: diffusion_policy.dataset.real_pusht_image_dataset.RealPushTImageDataset
|
|
35
|
+
shape_meta: *shape_meta
|
|
36
|
+
dataset_path: ${task.dataset_path}
|
|
37
|
+
horizon: ${horizon}
|
|
38
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
39
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
40
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
41
|
+
n_latency_steps: ${n_latency_steps}
|
|
42
|
+
use_cache: True
|
|
43
|
+
seed: 42
|
|
44
|
+
val_ratio: 0.00
|
|
45
|
+
max_train_episodes: null
|
|
46
|
+
delta_action: False
|
|
47
|
+
|
|
@@ -0,0 +1,72 @@
|
|
|
1
|
+
name: square_image
|
|
2
|
+
|
|
3
|
+
shape_meta: &shape_meta
|
|
4
|
+
# acceptable types: rgb, low_dim
|
|
5
|
+
obs:
|
|
6
|
+
agentview_image:
|
|
7
|
+
shape: [3, 84, 84]
|
|
8
|
+
horizon: 2
|
|
9
|
+
type: rgb
|
|
10
|
+
robot0_eye_in_hand_image:
|
|
11
|
+
shape: [3, 84, 84]
|
|
12
|
+
horizon: 2
|
|
13
|
+
type: rgb
|
|
14
|
+
robot0_eef_pos:
|
|
15
|
+
shape: [3]
|
|
16
|
+
horizon: 2
|
|
17
|
+
type: low_dim
|
|
18
|
+
robot0_eef_quat:
|
|
19
|
+
raw_shape: [4]
|
|
20
|
+
shape: [6]
|
|
21
|
+
horizon: 2
|
|
22
|
+
type: low_dim
|
|
23
|
+
rotation_rep: rotation_6d
|
|
24
|
+
robot0_gripper_qpos:
|
|
25
|
+
shape: [2]
|
|
26
|
+
horizon: 2
|
|
27
|
+
type: low_dim
|
|
28
|
+
action:
|
|
29
|
+
shape: [10]
|
|
30
|
+
rotation_rep: rotation_6d
|
|
31
|
+
|
|
32
|
+
task_name: &task_name square
|
|
33
|
+
dataset_type: &dataset_type ph
|
|
34
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/image_abs.hdf5
|
|
35
|
+
pose_repr: &pose_repr
|
|
36
|
+
obs_pose_repr: rel # abs or rel
|
|
37
|
+
action_pose_repr: rel
|
|
38
|
+
|
|
39
|
+
env_runner:
|
|
40
|
+
_target_: diffusion_policy.env_runner.robomimic_runner.RobomimicRunner
|
|
41
|
+
dataset_path: *dataset_path
|
|
42
|
+
pose_repr: *pose_repr
|
|
43
|
+
shape_meta: *shape_meta
|
|
44
|
+
n_train: 6
|
|
45
|
+
n_train_vis: 2
|
|
46
|
+
train_start_idx: 0
|
|
47
|
+
n_test: 50
|
|
48
|
+
n_test_vis: 4
|
|
49
|
+
test_start_seed: 100000
|
|
50
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
51
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
52
|
+
n_obs_steps: ${n_obs_steps}
|
|
53
|
+
n_action_steps: ${n_action_steps}
|
|
54
|
+
render_obs_key: 'agentview_image'
|
|
55
|
+
fps: 10
|
|
56
|
+
crf: 22
|
|
57
|
+
tqdm_interval_sec: 1.0
|
|
58
|
+
n_envs: 28
|
|
59
|
+
# evaluation at this config requires a 16 core 64GB instance.
|
|
60
|
+
|
|
61
|
+
dataset:
|
|
62
|
+
_target_: diffusion_policy.dataset.robomimic_replay_dataset.RobomimicReplayDataset
|
|
63
|
+
pose_repr: *pose_repr
|
|
64
|
+
shape_meta: *shape_meta
|
|
65
|
+
dataset_path: *dataset_path
|
|
66
|
+
horizon: ${horizon}
|
|
67
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
68
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
69
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
70
|
+
use_cache: True
|
|
71
|
+
seed: 42
|
|
72
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
name: square_image
|
|
2
|
+
|
|
3
|
+
shape_meta: &shape_meta
|
|
4
|
+
# acceptable types: rgb, low_dim
|
|
5
|
+
obs:
|
|
6
|
+
agentview_image:
|
|
7
|
+
shape: [3, 84, 84]
|
|
8
|
+
type: rgb
|
|
9
|
+
robot0_eye_in_hand_image:
|
|
10
|
+
shape: [3, 84, 84]
|
|
11
|
+
type: rgb
|
|
12
|
+
robot0_eef_pos:
|
|
13
|
+
shape: [3]
|
|
14
|
+
# type default: low_dim
|
|
15
|
+
robot0_eef_quat:
|
|
16
|
+
shape: [4]
|
|
17
|
+
robot0_gripper_qpos:
|
|
18
|
+
shape: [2]
|
|
19
|
+
action:
|
|
20
|
+
shape: [7]
|
|
21
|
+
|
|
22
|
+
task_name: &task_name square
|
|
23
|
+
dataset_type: &dataset_type ph
|
|
24
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/image.hdf5
|
|
25
|
+
abs_action: &abs_action False
|
|
26
|
+
|
|
27
|
+
env_runner:
|
|
28
|
+
_target_: diffusion_policy.env_runner.robomimic_image_runner.RobomimicImageRunner
|
|
29
|
+
dataset_path: *dataset_path
|
|
30
|
+
shape_meta: *shape_meta
|
|
31
|
+
# costs 1GB per env
|
|
32
|
+
n_train: 6
|
|
33
|
+
n_train_vis: 2
|
|
34
|
+
train_start_idx: 0
|
|
35
|
+
n_test: 50
|
|
36
|
+
n_test_vis: 4
|
|
37
|
+
test_start_seed: 100000
|
|
38
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
39
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
40
|
+
n_obs_steps: ${n_obs_steps}
|
|
41
|
+
n_action_steps: ${n_action_steps}
|
|
42
|
+
render_obs_key: 'agentview_image'
|
|
43
|
+
fps: 10
|
|
44
|
+
crf: 22
|
|
45
|
+
past_action: ${past_action_visible}
|
|
46
|
+
abs_action: *abs_action
|
|
47
|
+
tqdm_interval_sec: 1.0
|
|
48
|
+
n_envs: 28
|
|
49
|
+
# evaluation at this config requires a 16 core 64GB instance.
|
|
50
|
+
|
|
51
|
+
dataset:
|
|
52
|
+
_target_: diffusion_policy.dataset.robomimic_replay_image_dataset.RobomimicReplayImageDataset
|
|
53
|
+
shape_meta: *shape_meta
|
|
54
|
+
dataset_path: *dataset_path
|
|
55
|
+
horizon: ${horizon}
|
|
56
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
57
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
58
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
59
|
+
abs_action: *abs_action
|
|
60
|
+
rotation_rep: 'rotation_6d'
|
|
61
|
+
use_legacy_normalizer: False
|
|
62
|
+
use_cache: True
|
|
63
|
+
seed: 42
|
|
64
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
name: square_image
|
|
2
|
+
|
|
3
|
+
shape_meta: &shape_meta
|
|
4
|
+
# acceptable types: rgb, low_dim
|
|
5
|
+
obs:
|
|
6
|
+
agentview_image:
|
|
7
|
+
shape: [3, 84, 84]
|
|
8
|
+
type: rgb
|
|
9
|
+
robot0_eye_in_hand_image:
|
|
10
|
+
shape: [3, 84, 84]
|
|
11
|
+
type: rgb
|
|
12
|
+
robot0_eef_pos:
|
|
13
|
+
shape: [3]
|
|
14
|
+
# type default: low_dim
|
|
15
|
+
robot0_eef_quat:
|
|
16
|
+
shape: [4]
|
|
17
|
+
robot0_gripper_qpos:
|
|
18
|
+
shape: [2]
|
|
19
|
+
action:
|
|
20
|
+
shape: [10]
|
|
21
|
+
|
|
22
|
+
task_name: &task_name square
|
|
23
|
+
dataset_type: &dataset_type ph
|
|
24
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/image_abs.hdf5
|
|
25
|
+
abs_action: &abs_action True
|
|
26
|
+
|
|
27
|
+
env_runner:
|
|
28
|
+
_target_: diffusion_policy.env_runner.robomimic_image_runner.RobomimicImageRunner
|
|
29
|
+
dataset_path: *dataset_path
|
|
30
|
+
shape_meta: *shape_meta
|
|
31
|
+
# costs 1GB per env
|
|
32
|
+
n_train: 6
|
|
33
|
+
n_train_vis: 2
|
|
34
|
+
train_start_idx: 0
|
|
35
|
+
n_test: 50
|
|
36
|
+
n_test_vis: 4
|
|
37
|
+
test_start_seed: 100000
|
|
38
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
39
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
40
|
+
n_obs_steps: ${n_obs_steps}
|
|
41
|
+
n_action_steps: ${n_action_steps}
|
|
42
|
+
render_obs_key: 'agentview_image'
|
|
43
|
+
fps: 10
|
|
44
|
+
crf: 22
|
|
45
|
+
past_action: ${past_action_visible}
|
|
46
|
+
abs_action: *abs_action
|
|
47
|
+
tqdm_interval_sec: 1.0
|
|
48
|
+
n_envs: 28
|
|
49
|
+
# evaluation at this config requires a 16 core 64GB instance.
|
|
50
|
+
|
|
51
|
+
dataset:
|
|
52
|
+
_target_: diffusion_policy.dataset.robomimic_replay_image_dataset.RobomimicReplayImageDataset
|
|
53
|
+
shape_meta: *shape_meta
|
|
54
|
+
dataset_path: *dataset_path
|
|
55
|
+
horizon: ${horizon}
|
|
56
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
57
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
58
|
+
n_obs_steps: ${dataset_obs_steps}
|
|
59
|
+
abs_action: *abs_action
|
|
60
|
+
rotation_rep: 'rotation_6d'
|
|
61
|
+
use_legacy_normalizer: False
|
|
62
|
+
use_cache: True
|
|
63
|
+
seed: 42
|
|
64
|
+
val_ratio: 0.02
|
|
@@ -0,0 +1,46 @@
|
|
|
1
|
+
name: square_lowdim
|
|
2
|
+
|
|
3
|
+
obs_dim: 23
|
|
4
|
+
action_dim: 7
|
|
5
|
+
keypoint_dim: 3
|
|
6
|
+
|
|
7
|
+
obs_keys: &obs_keys ['object', 'robot0_eef_pos', 'robot0_eef_quat', 'robot0_gripper_qpos']
|
|
8
|
+
task_name: &task_name square
|
|
9
|
+
dataset_type: &dataset_type ph
|
|
10
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/low_dim.hdf5
|
|
11
|
+
abs_action: &abs_action False
|
|
12
|
+
|
|
13
|
+
env_runner:
|
|
14
|
+
_target_: diffusion_policy.env_runner.robomimic_lowdim_runner.RobomimicLowdimRunner
|
|
15
|
+
dataset_path: *dataset_path
|
|
16
|
+
obs_keys: *obs_keys
|
|
17
|
+
n_train: 6
|
|
18
|
+
n_train_vis: 2
|
|
19
|
+
train_start_idx: 0
|
|
20
|
+
n_test: 50
|
|
21
|
+
n_test_vis: 4
|
|
22
|
+
test_start_seed: 100000
|
|
23
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
24
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
25
|
+
n_obs_steps: ${n_obs_steps}
|
|
26
|
+
n_action_steps: ${n_action_steps}
|
|
27
|
+
n_latency_steps: ${n_latency_steps}
|
|
28
|
+
render_hw: [128,128]
|
|
29
|
+
fps: 10
|
|
30
|
+
crf: 22
|
|
31
|
+
past_action: ${past_action_visible}
|
|
32
|
+
abs_action: *abs_action
|
|
33
|
+
n_envs: 28
|
|
34
|
+
|
|
35
|
+
dataset:
|
|
36
|
+
_target_: diffusion_policy.dataset.robomimic_replay_lowdim_dataset.RobomimicReplayLowdimDataset
|
|
37
|
+
dataset_path: *dataset_path
|
|
38
|
+
horizon: ${horizon}
|
|
39
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
40
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
41
|
+
obs_keys: *obs_keys
|
|
42
|
+
abs_action: *abs_action
|
|
43
|
+
use_legacy_normalizer: False
|
|
44
|
+
seed: 42
|
|
45
|
+
val_ratio: 0.02
|
|
46
|
+
max_train_episodes: null
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
name: square_lowdim
|
|
2
|
+
|
|
3
|
+
obs_dim: 23
|
|
4
|
+
action_dim: 10
|
|
5
|
+
keypoint_dim: 3
|
|
6
|
+
|
|
7
|
+
obs_keys: &obs_keys ['object', 'robot0_eef_pos', 'robot0_eef_quat', 'robot0_gripper_qpos']
|
|
8
|
+
task_name: &task_name square
|
|
9
|
+
dataset_type: &dataset_type ph
|
|
10
|
+
abs_action: &abs_action True
|
|
11
|
+
dataset_path: &dataset_path data/robomimic/datasets/${task.task_name}/${task.dataset_type}/low_dim_abs.hdf5
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
env_runner:
|
|
15
|
+
_target_: diffusion_policy.env_runner.robomimic_lowdim_runner.RobomimicLowdimRunner
|
|
16
|
+
dataset_path: *dataset_path
|
|
17
|
+
obs_keys: *obs_keys
|
|
18
|
+
n_train: 6
|
|
19
|
+
n_train_vis: 2
|
|
20
|
+
train_start_idx: 0
|
|
21
|
+
n_test: 50
|
|
22
|
+
n_test_vis: 4
|
|
23
|
+
test_start_seed: 100000
|
|
24
|
+
# use python's eval function as resolver, single-quoted string as argument
|
|
25
|
+
max_steps: ${eval:'500 if "${task.dataset_type}" == "mh" else 400'}
|
|
26
|
+
n_obs_steps: ${n_obs_steps}
|
|
27
|
+
n_action_steps: ${n_action_steps}
|
|
28
|
+
n_latency_steps: ${n_latency_steps}
|
|
29
|
+
render_hw: [128,128]
|
|
30
|
+
fps: 10
|
|
31
|
+
crf: 22
|
|
32
|
+
past_action: ${past_action_visible}
|
|
33
|
+
abs_action: *abs_action
|
|
34
|
+
n_envs: 28
|
|
35
|
+
|
|
36
|
+
dataset:
|
|
37
|
+
_target_: diffusion_policy.dataset.robomimic_replay_lowdim_dataset.RobomimicReplayLowdimDataset
|
|
38
|
+
dataset_path: *dataset_path
|
|
39
|
+
horizon: ${horizon}
|
|
40
|
+
pad_before: ${eval:'${n_obs_steps}-1+${n_latency_steps}'}
|
|
41
|
+
pad_after: ${eval:'${horizon}-${n_obs_steps}'}
|
|
42
|
+
obs_keys: *obs_keys
|
|
43
|
+
abs_action: *abs_action
|
|
44
|
+
use_legacy_normalizer: False
|
|
45
|
+
seed: 42
|
|
46
|
+
val_ratio: 0.02
|
|
47
|
+
max_train_episodes: null
|