Hugging Face's logo

UFACTORY-Developer
/
ufactory_genesis_rl

Reinforcement Learning
rsl-rl
robotics
ppo
genesis
ufactory
xarm
manipulation
Model card Files
xet
 Community
UFACTORY-Developer commited on
Commit
c822bb3
·
verified ·
1 Parent(s): 7796c36

Upload Reach PPO checkpoints for xarm5, xarm6, xarm7, uf850, and lite6

Browse files
Files changed (26) hide show
  1. README.md +90 -0
  2. reach/lite6-reach/seed-7/acceptance-result.json +37 -0
  3. reach/lite6-reach/seed-7/checkpoint-selection.json +207 -0
  4. reach/lite6-reach/seed-7/config.yaml +147 -0
  5. reach/lite6-reach/seed-7/model_299.checkpoint_manifest.json +13 -0
  6. reach/lite6-reach/seed-7/model_299.pt +3 -0
  7. reach/uf850-reach/seed-7/acceptance-result.json +37 -0
  8. reach/uf850-reach/seed-7/checkpoint-selection.json +207 -0
  9. reach/uf850-reach/seed-7/config.yaml +147 -0
  10. reach/uf850-reach/seed-7/model_299.checkpoint_manifest.json +13 -0
  11. reach/uf850-reach/seed-7/model_299.pt +3 -0
  12. reach/xarm5-reach/seed-7/acceptance-result.json +37 -0
  13. reach/xarm5-reach/seed-7/checkpoint-selection.json +207 -0
  14. reach/xarm5-reach/seed-7/config.yaml +141 -0
  15. reach/xarm5-reach/seed-7/model_299.checkpoint_manifest.json +13 -0
  16. reach/xarm5-reach/seed-7/model_299.pt +3 -0
  17. reach/xarm6-reach/seed-7/acceptance-result.json +37 -0
  18. reach/xarm6-reach/seed-7/checkpoint-selection.json +207 -0
  19. reach/xarm6-reach/seed-7/config.yaml +147 -0
  20. reach/xarm6-reach/seed-7/model_299.checkpoint_manifest.json +13 -0
  21. reach/xarm6-reach/seed-7/model_299.pt +3 -0
  22. reach/xarm7-reach/seed-7/acceptance-result.json +37 -0
  23. reach/xarm7-reach/seed-7/checkpoint-selection.json +207 -0
  24. reach/xarm7-reach/seed-7/config.yaml +153 -0
  25. reach/xarm7-reach/seed-7/model_299.checkpoint_manifest.json +13 -0
  26. reach/xarm7-reach/seed-7/model_299.pt +3 -0
README.md ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ library_name: rsl-rl
4
+ tags:
5
+ - robotics
6
+ - reinforcement-learning
7
+ - ppo
8
+ - genesis
9
+ - ufactory
10
+ - xarm
11
+ - manipulation
12
+ ---
13
+
14
+ # UFACTORY Genesis Reinforcement Learning Checkpoints
15
+
16
+ Pretrained reinforcement learning policy checkpoints for [`ufactory_genesis`](https://github.com/xArm-Developer/ufactory_genesis), built on [Genesis](https://github.com/Genesis-Embodied-AI/Genesis) and `rsl-rl` PPO.
17
+
18
+ 基于 [Genesis](https://github.com/Genesis-Embodied-AI/Genesis) 与 `rsl-rl` PPO 训练的 [`ufactory_genesis`](https://github.com/xArm-Developer/ufactory_genesis) 强化学习预训练策略检查点。
19
+
20
+ ---
21
+
22
+ ## 1. Reach (`reach/`)
23
+
24
+ Bare-arm target reaching for 5 UFACTORY manipulators (`xarm5`, `xarm6`, `xarm7`, `uf850`, `lite6`):
25
+ - **Initial flange TCP pose**: `(300, 0, 300) mm`, `RPY = (180°, 0°, 0°)`
26
+ - **Success criterion**: strictly `< 5 mm` terminal distance (`success_distance_m = 0.005`)
27
+ - **Control rate**: `50 Hz` (`20 ms` step, `5 s` episode horizon), normalized joint-position increments (`[-1, 1]` -> max `0.01 rad/step`)
28
+ - **Training**: `rsl-rl` PPO (`[256, 256, 128]` ELU MLP with Beta distribution), `4096` parallel environments, `300` iterations, seed `7`
29
+
30
+ 5 款无夹爪 UFACTORY 机械臂（`xarm5`、`xarm6`、`xarm7`、`uf850`、`lite6`）目标到达策略：
31
+ - **初始法兰 TCP 位姿**：`(300, 0, 300) mm`，`RPY = (180°, 0°, 0°)`
32
+ - **成功判据**：末端与目标点距离严格小于 **5 mm**（`success_distance_m = 0.005`）
33
+ - **控制频率**：`50 Hz`（步长 `20 ms`，回合时长 `5 s`），归一化关节位置增量动作（每步最大 `0.01 rad`）
34
+ - **训练配置**：`rsl-rl` PPO（`[256, 256, 128]` ELU MLP，Beta 动作分布），`4096` 并行环境，`300` 轮迭代，种子 `7`
35
+
36
+ ### Benchmark & Holdout Validation / 验收指标
37
+
38
+ Evaluated on two independent holdout seeds (`17000` and `17001`, 512 random targets each):
39
+
40
+ | Robot (`--robot`) | DOF | Obs / Act Dim | Target Workspace (m) | Checkpoint | Holdout Success Rate (`17000` / `17001`) | Mean Terminal Distance (`17000` / `17001`) |
41
+ |---|---:|---:|---|---|---|---:|
42
+ | `xarm5` | 5 | 21 / 5 | x `0.15–0.55`, y `-0.30–0.30`, z `0.05–0.50` | `model_299.pt` | **99.6%** / **99.8%** | `4.48` / `4.50 mm` |
43
+ | `xarm6` | 6 | 24 / 6 | x `0.15–0.55`, y `-0.30–0.30`, z `0.05–0.50` | `model_299.pt` | **100%** / **100%** | `4.23` / `4.24 mm` |
44
+ | `xarm7` | 7 | 27 / 7 | x `0.15–0.55`, y `-0.30–0.30`, z `0.05–0.50` | `model_299.pt` | **99.8%** / **99.2%** | `4.39` / `4.52 mm` |
45
+ | `uf850` | 6 | 24 / 6 | x `0.15–0.55`, y `-0.30–0.30`, z `0.05–0.50` | `model_299.pt` | **99.6%** / **99.6%** | `4.26` / `4.34 mm` |
46
+ | `lite6` | 6 | 24 / 6 | x `0.15–0.35`, y `-0.20–0.20`, z `0.05–0.35` | `model_299.pt` | **99.8%** / **99.6%** | `4.66` / `4.71 mm` |
47
+
48
+ ---
49
+
50
+ ## Download & Usage / 下载与使用
51
+
52
+ Run from the root of the [`ufactory_genesis`](https://github.com/xArm-Developer/ufactory_genesis) repository:
53
+
54
+ 在 [`ufactory_genesis`](https://github.com/xArm-Developer/ufactory_genesis) 仓库根目录下执行：
55
+
56
+ ```bash
57
+ # Download all 5 Reach checkpoints into outputs/rl/reach/
58
+ # 下载全部 5 款机型的 Reach 检查点到 outputs/rl/reach/
59
+ hf download UFACTORY-Developer/ufactory_genesis_rl --include "reach/**" --local-dir outputs/rl
60
+
61
+ # Or download a single robot (e.g. xarm6)
62
+ # 或仅下载单个机型（以 xarm6 为例）
63
+ hf download UFACTORY-Developer/ufactory_genesis_rl --include "reach/xarm6-reach/**" --local-dir outputs/rl
64
+ ```
65
+
66
+ ### Real-time GLB Inference / GLB 实时可视化推理
67
+
68
+ ```bash
69
+ # Single arm / 单臂
70
+ python -m examples.rl.reach.visualize \
71
+ --checkpoint outputs/rl/reach/xarm6-reach/seed-7/model_299.pt
72
+
73
+ # Four arms / 四臂并行
74
+ python -m examples.rl.reach.visualize \
75
+ --checkpoint outputs/rl/reach/xarm6-reach/seed-7/model_299.pt -b 4
76
+ ```
77
+
78
+ ### Quantitative Evaluation / 定量评估
79
+
80
+ ```bash
81
+ python -m examples.rl.reach.evaluate \
82
+ --checkpoint outputs/rl/reach/xarm6-reach/seed-7/model_299.pt \
83
+ -B 128 --episodes 512 --seed 17000 \
84
+ --min-success-rate 0.95 --max-mean-distance-m 0.010 \
85
+ --quiet-episodes --summary-json outputs/rl/reach/xarm6-reach/seed-7/evaluation-17000.json
86
+ ```
87
+
88
+ See the full guides in the code repository:
89
+ - [English Guide (`examples/rl/reach/README.md`)](https://github.com/xArm-Developer/ufactory_genesis/blob/main/examples/rl/reach/README.md)
90
+ - [中文指南 (`examples/rl/reach/README_zh.md`)](https://github.com/xArm-Developer/ufactory_genesis/blob/main/examples/rl/reach/README_zh.md)
reach/lite6-reach/seed-7/acceptance-result.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_seed": 7,
3
+ "selected_checkpoint": "model_299.pt",
4
+ "passed": true,
5
+ "holdouts": [
6
+ {
7
+ "seed": 17000,
8
+ "summary": {
9
+ "episodes": 512,
10
+ "successes": 511,
11
+ "success_rate": 0.998046875,
12
+ "mean_m": 0.004663219671783736,
13
+ "median_m": 0.004673858638852835,
14
+ "p90_m": 0.0049403612967580555,
15
+ "p99_m": 0.004994685589335859,
16
+ "mean_elapsed_s": 1.3970703125000001,
17
+ "mean_success_elapsed_s": 1.3900195694716244
18
+ },
19
+ "report": "acceptance-17000.json"
20
+ },
21
+ {
22
+ "seed": 17001,
23
+ "summary": {
24
+ "episodes": 512,
25
+ "successes": 510,
26
+ "success_rate": 0.99609375,
27
+ "mean_m": 0.004709227724106313,
28
+ "median_m": 0.004664945881813765,
29
+ "p90_m": 0.004935949109494686,
30
+ "p99_m": 0.0049942695070058106,
31
+ "mean_elapsed_s": 1.4323828125,
32
+ "mean_success_elapsed_s": 1.4183921568627451
33
+ },
34
+ "report": "acceptance-17001.json"
35
+ }
36
+ ]
37
+ }
reach/lite6-reach/seed-7/checkpoint-selection.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "selected_checkpoint": "model_299.pt",
3
+ "candidates": [
4
+ {
5
+ "checkpoint": "model_299.pt",
6
+ "report": "model_299-development.json",
7
+ "summary": {
8
+ "episodes": 128,
9
+ "successes": 128,
10
+ "success_rate": 1.0,
11
+ "mean_m": 0.0045977348290762166,
12
+ "median_m": 0.0046960776671767235,
13
+ "p90_m": 0.004952587746083736,
14
+ "p99_m": 0.0049920148123055695,
15
+ "mean_elapsed_s": 1.40703125,
16
+ "mean_success_elapsed_s": 1.40703125
17
+ }
18
+ },
19
+ {
20
+ "checkpoint": "model_275.pt",
21
+ "report": "model_275-development.json",
22
+ "summary": {
23
+ "episodes": 128,
24
+ "successes": 126,
25
+ "success_rate": 0.984375,
26
+ "mean_m": 0.00496404755540425,
27
+ "median_m": 0.004723179619759321,
28
+ "p90_m": 0.004959360975772142,
29
+ "p99_m": 0.01673933062702424,
30
+ "mean_elapsed_s": 1.55484375,
31
+ "mean_success_elapsed_s": 1.5001587301587302
32
+ }
33
+ },
34
+ {
35
+ "checkpoint": "model_250.pt",
36
+ "report": "model_250-development.json",
37
+ "summary": {
38
+ "episodes": 128,
39
+ "successes": 124,
40
+ "success_rate": 0.96875,
41
+ "mean_m": 0.005242108787570032,
42
+ "median_m": 0.0048275848384946585,
43
+ "p90_m": 0.00498148831538856,
44
+ "p99_m": 0.022732150368392483,
45
+ "mean_elapsed_s": 1.8223437500000002,
46
+ "mean_success_elapsed_s": 1.7198387096774195
47
+ }
48
+ },
49
+ {
50
+ "checkpoint": "model_225.pt",
51
+ "report": "model_225-development.json",
52
+ "summary": {
53
+ "episodes": 128,
54
+ "successes": 105,
55
+ "success_rate": 0.8203125,
56
+ "mean_m": 0.0080501918491791,
57
+ "median_m": 0.004893379285931587,
58
+ "p90_m": 0.021784577704966064,
59
+ "p99_m": 0.03378026466816664,
60
+ "mean_elapsed_s": 2.5090625,
61
+ "mean_success_elapsed_s": 1.9634285714285713
62
+ }
63
+ },
64
+ {
65
+ "checkpoint": "model_200.pt",
66
+ "report": "model_200-development.json",
67
+ "summary": {
68
+ "episodes": 128,
69
+ "successes": 80,
70
+ "success_rate": 0.625,
71
+ "mean_m": 0.012298004414333263,
72
+ "median_m": 0.004961335798725486,
73
+ "p90_m": 0.028689396381378175,
74
+ "p99_m": 0.059081536717712886,
75
+ "mean_elapsed_s": 3.23,
76
+ "mean_success_elapsed_s": 2.168
77
+ }
78
+ },
79
+ {
80
+ "checkpoint": "model_175.pt",
81
+ "report": "model_175-development.json",
82
+ "summary": {
83
+ "episodes": 128,
84
+ "successes": 38,
85
+ "success_rate": 0.296875,
86
+ "mean_m": 0.01835856608886388,
87
+ "median_m": 0.015913539566099644,
88
+ "p90_m": 0.03723196983337402,
89
+ "p99_m": 0.05129575125873089,
90
+ "mean_elapsed_s": 4.281875,
91
+ "mean_success_elapsed_s": 2.5810526315789475
92
+ }
93
+ },
94
+ {
95
+ "checkpoint": "model_150.pt",
96
+ "report": "model_150-development.json",
97
+ "summary": {
98
+ "episodes": 128,
99
+ "successes": 16,
100
+ "success_rate": 0.125,
101
+ "mean_m": 0.022742844055755995,
102
+ "median_m": 0.018767801113426685,
103
+ "p90_m": 0.046165432035923,
104
+ "p99_m": 0.08371392585337163,
105
+ "mean_elapsed_s": 4.7153125,
106
+ "mean_success_elapsed_s": 2.7225
107
+ }
108
+ },
109
+ {
110
+ "checkpoint": "model_125.pt",
111
+ "report": "model_125-development.json",
112
+ "summary": {
113
+ "episodes": 128,
114
+ "successes": 9,
115
+ "success_rate": 0.0703125,
116
+ "mean_m": 0.044236479840037646,
117
+ "median_m": 0.03295031934976578,
118
+ "p90_m": 0.09410616531968115,
119
+ "p99_m": 0.1310846158862114,
120
+ "mean_elapsed_s": 4.8646875,
121
+ "mean_success_elapsed_s": 3.0755555555555554
122
+ }
123
+ },
124
+ {
125
+ "checkpoint": "model_100.pt",
126
+ "report": "model_100-development.json",
127
+ "summary": {
128
+ "episodes": 128,
129
+ "successes": 1,
130
+ "success_rate": 0.0078125,
131
+ "mean_m": 0.07486774368953775,
132
+ "median_m": 0.0694241113960743,
133
+ "p90_m": 0.12722332775592804,
134
+ "p99_m": 0.24588850870728496,
135
+ "mean_elapsed_s": 4.969375,
136
+ "mean_success_elapsed_s": 1.08
137
+ }
138
+ },
139
+ {
140
+ "checkpoint": "model_75.pt",
141
+ "report": "model_75-development.json",
142
+ "summary": {
143
+ "episodes": 128,
144
+ "successes": 0,
145
+ "success_rate": 0.0,
146
+ "mean_m": 0.14432954203221016,
147
+ "median_m": 0.13988403230905533,
148
+ "p90_m": 0.2370660200715065,
149
+ "p99_m": 0.2722295045852661,
150
+ "mean_elapsed_s": 5.0,
151
+ "mean_success_elapsed_s": null
152
+ }
153
+ },
154
+ {
155
+ "checkpoint": "model_25.pt",
156
+ "report": "model_25-development.json",
157
+ "summary": {
158
+ "episodes": 128,
159
+ "successes": 0,
160
+ "success_rate": 0.0,
161
+ "mean_m": 0.14746090312837623,
162
+ "median_m": 0.1481284573674202,
163
+ "p90_m": 0.20736985355615614,
164
+ "p99_m": 0.2429337152838707,
165
+ "mean_elapsed_s": 5.0,
166
+ "mean_success_elapsed_s": null
167
+ }
168
+ },
169
+ {
170
+ "checkpoint": "model_50.pt",
171
+ "report": "model_50-development.json",
172
+ "summary": {
173
+ "episodes": 128,
174
+ "successes": 0,
175
+ "success_rate": 0.0,
176
+ "mean_m": 0.155796767125139,
177
+ "median_m": 0.15427003055810928,
178
+ "p90_m": 0.2184711143374443,
179
+ "p99_m": 0.26542443782091146,
180
+ "mean_elapsed_s": 5.0,
181
+ "mean_success_elapsed_s": null
182
+ }
183
+ },
184
+ {
185
+ "checkpoint": "model_0.pt",
186
+ "report": "model_0-development.json",
187
+ "summary": {
188
+ "episodes": 128,
189
+ "successes": 0,
190
+ "success_rate": 0.0,
191
+ "mean_m": 0.16346461261127843,
192
+ "median_m": 0.16769902408123016,
193
+ "p90_m": 0.23155205994844436,
194
+ "p99_m": 0.2644365760684014,
195
+ "mean_elapsed_s": 5.0,
196
+ "mean_success_elapsed_s": null
197
+ }
198
+ }
199
+ ],
200
+ "development_seed": 17002,
201
+ "development_episodes": 128,
202
+ "ranking": [
203
+ "success_rate_desc",
204
+ "mean_m_asc",
205
+ "mean_success_elapsed_s_asc"
206
+ ]
207
+ }
reach/lite6-reach/seed-7/config.yaml ADDED
@@ -0,0 +1,147 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 1
2
+ task: reach
3
+ robot_key: lite6
4
+ env:
5
+ num_envs: 4096
6
+ genesis_performance_mode: true
7
+ randomize_initial_episode_length: true
8
+ task_name: reach
9
+ env_version: 2
10
+ num_obs: 24
11
+ num_actions: 6
12
+ runtime_config_sha256: f17e703be7ac907ce502ba2c8bb3815761f8672d5b55017a00c825159e198302
13
+ episode_length_s: 5.0
14
+ ctrl_dt: 0.02
15
+ success_distance_m: 0.005
16
+ action_scale: 0.05
17
+ action_clip: 1.0
18
+ servo_speed_rad_s: 0.5
19
+ max_joint_delta_rad: 0.01
20
+ target_pos_lower:
21
+ - 0.15
22
+ - -0.2
23
+ - 0.05
24
+ target_pos_upper:
25
+ - 0.35
26
+ - 0.2
27
+ - 0.35
28
+ substeps: 32
29
+ constraint_solver: newton
30
+ solver_iterations: 100
31
+ noslip_iterations: 0
32
+ friction_cone: pyramidal
33
+ contact_resolution: convex
34
+ constraint_time_constant_s: 0.005
35
+ use_gjk_collision: true
36
+ reward:
37
+ reach: 1.0
38
+ action_penalty: 0.001
39
+ success: 2.0
40
+ robot:
41
+ urdf_path: assets/urdf/lite6/lite6.urdf
42
+ base_pos:
43
+ - 0.0
44
+ - 0.0
45
+ - 0.0
46
+ ee_link_name: link6
47
+ joint_names:
48
+ - joint1
49
+ - joint2
50
+ - joint3
51
+ - joint4
52
+ - joint5
53
+ - joint6
54
+ default_qpos:
55
+ - 0.0
56
+ - 0.338839
57
+ - 1.253953
58
+ - 0.0
59
+ - 0.915106
60
+ - 0.0
61
+ kp:
62
+ - 3000.0
63
+ - 3000.0
64
+ - 2000.0
65
+ - 40.0
66
+ - 80.0
67
+ - 40.0
68
+ kv:
69
+ - 300.0
70
+ - 300.0
71
+ - 200.0
72
+ - 4.0
73
+ - 8.0
74
+ - 4.0
75
+ force_lower:
76
+ - -81.6
77
+ - -81.6
78
+ - -30.6
79
+ - -1.364505
80
+ - -3.274812
81
+ - -1.364505
82
+ force_upper:
83
+ - 81.6
84
+ - 81.6
85
+ - 30.6
86
+ - 1.364505
87
+ - 3.274812
88
+ - 1.364505
89
+ train:
90
+ algorithm:
91
+ class_name: PPO
92
+ clip_param: 0.2
93
+ desired_kl: 0.01
94
+ entropy_coef: 0.001
95
+ gamma: 0.99
96
+ lam: 0.95
97
+ learning_rate: 0.0003
98
+ max_grad_norm: 1.0
99
+ num_learning_epochs: 5
100
+ num_mini_batches: 4
101
+ schedule: fixed
102
+ use_clipped_value_loss: true
103
+ value_loss_coef: 1.0
104
+ rnd_cfg: null
105
+ symmetry_cfg: null
106
+ actor:
107
+ class_name: MLPModel
108
+ hidden_dims:
109
+ - 256
110
+ - 256
111
+ - 128
112
+ activation: elu
113
+ obs_normalization: true
114
+ distribution_cfg:
115
+ class_name: BetaDistribution
116
+ action_range:
117
+ - -1.0
118
+ - 1.0
119
+ critic:
120
+ class_name: MLPModel
121
+ hidden_dims:
122
+ - 256
123
+ - 256
124
+ - 128
125
+ activation: elu
126
+ obs_normalization: true
127
+ obs_groups:
128
+ actor:
129
+ - policy
130
+ critic:
131
+ - policy
132
+ runner:
133
+ checkpoint: -1
134
+ experiment_name: lite6-reach
135
+ load_run: -1
136
+ log_interval: 1
137
+ max_iterations: 300
138
+ record_interval: -1
139
+ resume: false
140
+ resume_path: null
141
+ run_name: ''
142
+ num_steps_per_env: 24
143
+ save_interval: 25
144
+ seed: 7
145
+ run_name: ''
146
+ logger: tensorboard
147
+ config_sha256: 2dbe418c335e3216d3f526b0cbf9d122d501415231b8f389ae97e5be1416e40b
reach/lite6-reach/seed-7/model_299.checkpoint_manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 6,
3
+ "checkpoint_file": "model_299.pt",
4
+ "checkpoint_sha256": "e60eeccd2557651886571a1a0bf4d1008cabd765a9a1ba698fe1a26cb3797444",
5
+ "code_version": "c4aaa345ec0d702ae7801373965c0b5ccf5f7576",
6
+ "config_sha256": "2dbe418c335e3216d3f526b0cbf9d122d501415231b8f389ae97e5be1416e40b",
7
+ "dof": 6,
8
+ "executor_action_contract": "normalized_joint_position_delta",
9
+ "observation_dim": 24,
10
+ "robot_key": "lite6",
11
+ "schema_version": 1,
12
+ "task": "reach"
13
+ }
reach/lite6-reach/seed-7/model_299.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e60eeccd2557651886571a1a0bf4d1008cabd765a9a1ba698fe1a26cb3797444
3
+ size 3027036
reach/uf850-reach/seed-7/acceptance-result.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_seed": 7,
3
+ "selected_checkpoint": "model_299.pt",
4
+ "passed": true,
5
+ "holdouts": [
6
+ {
7
+ "seed": 17000,
8
+ "summary": {
9
+ "episodes": 512,
10
+ "successes": 510,
11
+ "success_rate": 0.99609375,
12
+ "mean_m": 0.0042641937698135735,
13
+ "median_m": 0.004344700137153268,
14
+ "p90_m": 0.004893100447952747,
15
+ "p99_m": 0.004992876038886607,
16
+ "mean_elapsed_s": 1.368203125,
17
+ "mean_success_elapsed_s": 1.3539607843137254
18
+ },
19
+ "report": "acceptance-17000.json"
20
+ },
21
+ {
22
+ "seed": 17001,
23
+ "summary": {
24
+ "episodes": 512,
25
+ "successes": 510,
26
+ "success_rate": 0.99609375,
27
+ "mean_m": 0.004340191737810528,
28
+ "median_m": 0.004415665054693818,
29
+ "p90_m": 0.004887049226090312,
30
+ "p99_m": 0.0049956104485318065,
31
+ "mean_elapsed_s": 1.3681640625,
32
+ "mean_success_elapsed_s": 1.353921568627451
33
+ },
34
+ "report": "acceptance-17001.json"
35
+ }
36
+ ]
37
+ }
reach/uf850-reach/seed-7/checkpoint-selection.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "selected_checkpoint": "model_299.pt",
3
+ "candidates": [
4
+ {
5
+ "checkpoint": "model_299.pt",
6
+ "report": "model_299-development.json",
7
+ "summary": {
8
+ "episodes": 128,
9
+ "successes": 128,
10
+ "success_rate": 1.0,
11
+ "mean_m": 0.004294230320738279,
12
+ "median_m": 0.004429191583767533,
13
+ "p90_m": 0.00491092475131154,
14
+ "p99_m": 0.004995538317598402,
15
+ "mean_elapsed_s": 1.34671875,
16
+ "mean_success_elapsed_s": 1.34671875
17
+ }
18
+ },
19
+ {
20
+ "checkpoint": "model_275.pt",
21
+ "report": "model_275-development.json",
22
+ "summary": {
23
+ "episodes": 128,
24
+ "successes": 128,
25
+ "success_rate": 1.0,
26
+ "mean_m": 0.004380268977911328,
27
+ "median_m": 0.00457098288461566,
28
+ "p90_m": 0.004931127745658159,
29
+ "p99_m": 0.004995037838816643,
30
+ "mean_elapsed_s": 1.3804687500000001,
31
+ "mean_success_elapsed_s": 1.3804687500000001
32
+ }
33
+ },
34
+ {
35
+ "checkpoint": "model_250.pt",
36
+ "report": "model_250-development.json",
37
+ "summary": {
38
+ "episodes": 128,
39
+ "successes": 128,
40
+ "success_rate": 1.0,
41
+ "mean_m": 0.004459714999029529,
42
+ "median_m": 0.004531538346782327,
43
+ "p90_m": 0.004909787699580193,
44
+ "p99_m": 0.004992926414124668,
45
+ "mean_elapsed_s": 1.42328125,
46
+ "mean_success_elapsed_s": 1.42328125
47
+ }
48
+ },
49
+ {
50
+ "checkpoint": "model_225.pt",
51
+ "report": "model_225-development.json",
52
+ "summary": {
53
+ "episodes": 128,
54
+ "successes": 126,
55
+ "success_rate": 0.984375,
56
+ "mean_m": 0.004788097230630228,
57
+ "median_m": 0.004620962776243687,
58
+ "p90_m": 0.004937815805897116,
59
+ "p99_m": 0.014702307195402735,
60
+ "mean_elapsed_s": 1.570625,
61
+ "mean_success_elapsed_s": 1.516190476190476
62
+ }
63
+ },
64
+ {
65
+ "checkpoint": "model_175.pt",
66
+ "report": "model_175-development.json",
67
+ "summary": {
68
+ "episodes": 128,
69
+ "successes": 116,
70
+ "success_rate": 0.90625,
71
+ "mean_m": 0.006642140497206128,
72
+ "median_m": 0.004768101032823324,
73
+ "p90_m": 0.004988536657765508,
74
+ "p99_m": 0.04644740816205743,
75
+ "mean_elapsed_s": 2.04046875,
76
+ "mean_success_elapsed_s": 1.7343103448275863
77
+ }
78
+ },
79
+ {
80
+ "checkpoint": "model_200.pt",
81
+ "report": "model_200-development.json",
82
+ "summary": {
83
+ "episodes": 128,
84
+ "successes": 116,
85
+ "success_rate": 0.90625,
86
+ "mean_m": 0.007073425264024991,
87
+ "median_m": 0.00473488075658679,
88
+ "p90_m": 0.004996079113334417,
89
+ "p99_m": 0.05189259346574547,
90
+ "mean_elapsed_s": 1.920625,
91
+ "mean_success_elapsed_s": 1.6020689655172413
92
+ }
93
+ },
94
+ {
95
+ "checkpoint": "model_150.pt",
96
+ "report": "model_150-development.json",
97
+ "summary": {
98
+ "episodes": 128,
99
+ "successes": 108,
100
+ "success_rate": 0.84375,
101
+ "mean_m": 0.006686581917165313,
102
+ "median_m": 0.004870233125984669,
103
+ "p90_m": 0.011884229350835083,
104
+ "p99_m": 0.030268666241318007,
105
+ "mean_elapsed_s": 2.5,
106
+ "mean_success_elapsed_s": 2.037037037037037
107
+ }
108
+ },
109
+ {
110
+ "checkpoint": "model_125.pt",
111
+ "report": "model_125-development.json",
112
+ "summary": {
113
+ "episodes": 128,
114
+ "successes": 74,
115
+ "success_rate": 0.578125,
116
+ "mean_m": 0.009518064438452711,
117
+ "median_m": 0.004989576293155551,
118
+ "p90_m": 0.02007368002086878,
119
+ "p99_m": 0.03660471815615893,
120
+ "mean_elapsed_s": 3.668125,
121
+ "mean_success_elapsed_s": 2.696216216216216
122
+ }
123
+ },
124
+ {
125
+ "checkpoint": "model_100.pt",
126
+ "report": "model_100-development.json",
127
+ "summary": {
128
+ "episodes": 128,
129
+ "successes": 6,
130
+ "success_rate": 0.046875,
131
+ "mean_m": 0.04108818856548169,
132
+ "median_m": 0.040397388860583305,
133
+ "p90_m": 0.06618981435894966,
134
+ "p99_m": 0.09724268302321434,
135
+ "mean_elapsed_s": 4.90375,
136
+ "mean_success_elapsed_s": 2.9466666666666668
137
+ }
138
+ },
139
+ {
140
+ "checkpoint": "model_75.pt",
141
+ "report": "model_75-development.json",
142
+ "summary": {
143
+ "episodes": 128,
144
+ "successes": 1,
145
+ "success_rate": 0.0078125,
146
+ "mean_m": 0.10104072008834919,
147
+ "median_m": 0.0874025896191597,
148
+ "p90_m": 0.17400038689374925,
149
+ "p99_m": 0.26876385420560844,
150
+ "mean_elapsed_s": 4.9875,
151
+ "mean_success_elapsed_s": 3.4
152
+ }
153
+ },
154
+ {
155
+ "checkpoint": "model_50.pt",
156
+ "report": "model_50-development.json",
157
+ "summary": {
158
+ "episodes": 128,
159
+ "successes": 1,
160
+ "success_rate": 0.0078125,
161
+ "mean_m": 0.1642930331399839,
162
+ "median_m": 0.12952835857868195,
163
+ "p90_m": 0.28327473998069747,
164
+ "p99_m": 0.6135977303981781,
165
+ "mean_elapsed_s": 4.98484375,
166
+ "mean_success_elapsed_s": 3.06
167
+ }
168
+ },
169
+ {
170
+ "checkpoint": "model_25.pt",
171
+ "report": "model_25-development.json",
172
+ "summary": {
173
+ "episodes": 128,
174
+ "successes": 0,
175
+ "success_rate": 0.0,
176
+ "mean_m": 0.23320347751723602,
177
+ "median_m": 0.2188900262117386,
178
+ "p90_m": 0.37299002408981324,
179
+ "p99_m": 0.5149619942903518,
180
+ "mean_elapsed_s": 5.0,
181
+ "mean_success_elapsed_s": null
182
+ }
183
+ },
184
+ {
185
+ "checkpoint": "model_0.pt",
186
+ "report": "model_0-development.json",
187
+ "summary": {
188
+ "episodes": 128,
189
+ "successes": 0,
190
+ "success_rate": 0.0,
191
+ "mean_m": 0.26139233011053875,
192
+ "median_m": 0.2743360996246338,
193
+ "p90_m": 0.3684306085109711,
194
+ "p99_m": 0.4168894076347352,
195
+ "mean_elapsed_s": 5.0,
196
+ "mean_success_elapsed_s": null
197
+ }
198
+ }
199
+ ],
200
+ "development_seed": 17002,
201
+ "development_episodes": 128,
202
+ "ranking": [
203
+ "success_rate_desc",
204
+ "mean_m_asc",
205
+ "mean_success_elapsed_s_asc"
206
+ ]
207
+ }
reach/uf850-reach/seed-7/config.yaml ADDED
@@ -0,0 +1,147 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 1
2
+ task: reach
3
+ robot_key: uf850
4
+ env:
5
+ num_envs: 4096
6
+ genesis_performance_mode: true
7
+ randomize_initial_episode_length: true
8
+ task_name: reach
9
+ env_version: 2
10
+ num_obs: 24
11
+ num_actions: 6
12
+ runtime_config_sha256: d03354e9ae37de078b0cf0701183335278f54417ee6a487258b71b7d9e5ab885
13
+ episode_length_s: 5.0
14
+ ctrl_dt: 0.02
15
+ success_distance_m: 0.005
16
+ action_scale: 0.05
17
+ action_clip: 1.0
18
+ servo_speed_rad_s: 0.5
19
+ max_joint_delta_rad: 0.01
20
+ target_pos_lower:
21
+ - 0.15
22
+ - -0.3
23
+ - 0.05
24
+ target_pos_upper:
25
+ - 0.55
26
+ - 0.3
27
+ - 0.5
28
+ substeps: 32
29
+ constraint_solver: newton
30
+ solver_iterations: 100
31
+ noslip_iterations: 0
32
+ friction_cone: pyramidal
33
+ contact_resolution: convex
34
+ constraint_time_constant_s: 0.00125
35
+ use_gjk_collision: true
36
+ reward:
37
+ reach: 1.0
38
+ action_penalty: 0.001
39
+ success: 2.0
40
+ robot:
41
+ urdf_path: assets/urdf/uf850/uf850.urdf
42
+ base_pos:
43
+ - 0.0
44
+ - 0.0
45
+ - 0.0
46
+ ee_link_name: link6
47
+ joint_names:
48
+ - joint1
49
+ - joint2
50
+ - joint3
51
+ - joint4
52
+ - joint5
53
+ - joint6
54
+ default_qpos:
55
+ - 0.0
56
+ - -0.079506
57
+ - -0.37905
58
+ - 0.0
59
+ - -0.299544
60
+ - 0.0
61
+ kp:
62
+ - 3000.0
63
+ - 3000.0
64
+ - 2000.0
65
+ - 2000.0
66
+ - 1000.0
67
+ - 1000.0
68
+ kv:
69
+ - 300.0
70
+ - 300.0
71
+ - 200.0
72
+ - 200.0
73
+ - 100.0
74
+ - 100.0
75
+ force_lower:
76
+ - -382.5
77
+ - -382.5
78
+ - -243.525
79
+ - -117.3
80
+ - -38.25
81
+ - -38.25
82
+ force_upper:
83
+ - 382.5
84
+ - 382.5
85
+ - 243.525
86
+ - 117.3
87
+ - 38.25
88
+ - 38.25
89
+ train:
90
+ algorithm:
91
+ class_name: PPO
92
+ clip_param: 0.2
93
+ desired_kl: 0.01
94
+ entropy_coef: 0.001
95
+ gamma: 0.99
96
+ lam: 0.95
97
+ learning_rate: 0.0003
98
+ max_grad_norm: 1.0
99
+ num_learning_epochs: 5
100
+ num_mini_batches: 4
101
+ schedule: fixed
102
+ use_clipped_value_loss: true
103
+ value_loss_coef: 1.0
104
+ rnd_cfg: null
105
+ symmetry_cfg: null
106
+ actor:
107
+ class_name: MLPModel
108
+ hidden_dims:
109
+ - 256
110
+ - 256
111
+ - 128
112
+ activation: elu
113
+ obs_normalization: true
114
+ distribution_cfg:
115
+ class_name: BetaDistribution
116
+ action_range:
117
+ - -1.0
118
+ - 1.0
119
+ critic:
120
+ class_name: MLPModel
121
+ hidden_dims:
122
+ - 256
123
+ - 256
124
+ - 128
125
+ activation: elu
126
+ obs_normalization: true
127
+ obs_groups:
128
+ actor:
129
+ - policy
130
+ critic:
131
+ - policy
132
+ runner:
133
+ checkpoint: -1
134
+ experiment_name: uf850-reach
135
+ load_run: -1
136
+ log_interval: 1
137
+ max_iterations: 300
138
+ record_interval: -1
139
+ resume: false
140
+ resume_path: null
141
+ run_name: ''
142
+ num_steps_per_env: 24
143
+ save_interval: 25
144
+ seed: 7
145
+ run_name: ''
146
+ logger: tensorboard
147
+ config_sha256: 46ee2999dd31f9f0e162523082f804fc87e66ca18a2bcc084abcbed7cae6f729
reach/uf850-reach/seed-7/model_299.checkpoint_manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 6,
3
+ "checkpoint_file": "model_299.pt",
4
+ "checkpoint_sha256": "bc91b7bd0ecfdf33d1856f050b63dfdc3ff3f22c9668abfb10b62acd3a60c322",
5
+ "code_version": "c4aaa345ec0d702ae7801373965c0b5ccf5f7576",
6
+ "config_sha256": "46ee2999dd31f9f0e162523082f804fc87e66ca18a2bcc084abcbed7cae6f729",
7
+ "dof": 6,
8
+ "executor_action_contract": "normalized_joint_position_delta",
9
+ "observation_dim": 24,
10
+ "robot_key": "uf850",
11
+ "schema_version": 1,
12
+ "task": "reach"
13
+ }
reach/uf850-reach/seed-7/model_299.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc91b7bd0ecfdf33d1856f050b63dfdc3ff3f22c9668abfb10b62acd3a60c322
3
+ size 3027036
reach/xarm5-reach/seed-7/acceptance-result.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_seed": 7,
3
+ "selected_checkpoint": "model_299.pt",
4
+ "passed": true,
5
+ "holdouts": [
6
+ {
7
+ "seed": 17000,
8
+ "summary": {
9
+ "episodes": 512,
10
+ "successes": 510,
11
+ "success_rate": 0.99609375,
12
+ "mean_m": 0.004483080940644868,
13
+ "median_m": 0.00457295379601419,
14
+ "p90_m": 0.004945111647248268,
15
+ "p99_m": 0.0049965945072472095,
16
+ "mean_elapsed_s": 1.7423046875000001,
17
+ "mean_success_elapsed_s": 1.729529411764706
18
+ },
19
+ "report": "acceptance-17000.json"
20
+ },
21
+ {
22
+ "seed": 17001,
23
+ "summary": {
24
+ "episodes": 512,
25
+ "successes": 511,
26
+ "success_rate": 0.998046875,
27
+ "mean_m": 0.004497742240801017,
28
+ "median_m": 0.004633271135389805,
29
+ "p90_m": 0.004947336157783866,
30
+ "p99_m": 0.004996288185939193,
31
+ "mean_elapsed_s": 1.721015625,
32
+ "mean_success_elapsed_s": 1.7145988258317024
33
+ },
34
+ "report": "acceptance-17001.json"
35
+ }
36
+ ]
37
+ }
reach/xarm5-reach/seed-7/checkpoint-selection.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "selected_checkpoint": "model_299.pt",
3
+ "candidates": [
4
+ {
5
+ "checkpoint": "model_299.pt",
6
+ "report": "model_299-development.json",
7
+ "summary": {
8
+ "episodes": 128,
9
+ "successes": 128,
10
+ "success_rate": 1.0,
11
+ "mean_m": 0.004440336520929122,
12
+ "median_m": 0.004522846778854728,
13
+ "p90_m": 0.004958468768745661,
14
+ "p99_m": 0.004996130270883441,
15
+ "mean_elapsed_s": 1.7353125,
16
+ "mean_success_elapsed_s": 1.7353125
17
+ }
18
+ },
19
+ {
20
+ "checkpoint": "model_250.pt",
21
+ "report": "model_250-development.json",
22
+ "summary": {
23
+ "episodes": 128,
24
+ "successes": 128,
25
+ "success_rate": 1.0,
26
+ "mean_m": 0.004568720836687135,
27
+ "median_m": 0.004657628946006298,
28
+ "p90_m": 0.004937981860712171,
29
+ "p99_m": 0.004993949513882399,
30
+ "mean_elapsed_s": 1.81703125,
31
+ "mean_success_elapsed_s": 1.81703125
32
+ }
33
+ },
34
+ {
35
+ "checkpoint": "model_225.pt",
36
+ "report": "model_225-development.json",
37
+ "summary": {
38
+ "episodes": 128,
39
+ "successes": 128,
40
+ "success_rate": 1.0,
41
+ "mean_m": 0.004600814911100315,
42
+ "median_m": 0.0047276923432946205,
43
+ "p90_m": 0.004937690915539861,
44
+ "p99_m": 0.004991318858228624,
45
+ "mean_elapsed_s": 1.8987500000000002,
46
+ "mean_success_elapsed_s": 1.8987500000000002
47
+ }
48
+ },
49
+ {
50
+ "checkpoint": "model_275.pt",
51
+ "report": "model_275-development.json",
52
+ "summary": {
53
+ "episodes": 128,
54
+ "successes": 127,
55
+ "success_rate": 0.9921875,
56
+ "mean_m": 0.004609594832800212,
57
+ "median_m": 0.0047398589085787535,
58
+ "p90_m": 0.004961234517395496,
59
+ "p99_m": 0.004998069358989597,
60
+ "mean_elapsed_s": 1.80765625,
61
+ "mean_success_elapsed_s": 1.78251968503937
62
+ }
63
+ },
64
+ {
65
+ "checkpoint": "model_200.pt",
66
+ "report": "model_200-development.json",
67
+ "summary": {
68
+ "episodes": 128,
69
+ "successes": 124,
70
+ "success_rate": 0.96875,
71
+ "mean_m": 0.005458343139252975,
72
+ "median_m": 0.004808088298887014,
73
+ "p90_m": 0.004982767719775438,
74
+ "p99_m": 0.016194920707494023,
75
+ "mean_elapsed_s": 2.1875,
76
+ "mean_success_elapsed_s": 2.096774193548387
77
+ }
78
+ },
79
+ {
80
+ "checkpoint": "model_175.pt",
81
+ "report": "model_175-development.json",
82
+ "summary": {
83
+ "episodes": 128,
84
+ "successes": 115,
85
+ "success_rate": 0.8984375,
86
+ "mean_m": 0.006561983878782485,
87
+ "median_m": 0.004795112181454897,
88
+ "p90_m": 0.005839618900790803,
89
+ "p99_m": 0.04632855884730822,
90
+ "mean_elapsed_s": 2.5828125,
91
+ "mean_success_elapsed_s": 2.3095652173913046
92
+ }
93
+ },
94
+ {
95
+ "checkpoint": "model_150.pt",
96
+ "report": "model_150-development.json",
97
+ "summary": {
98
+ "episodes": 128,
99
+ "successes": 58,
100
+ "success_rate": 0.453125,
101
+ "mean_m": 0.017583969645784236,
102
+ "median_m": 0.012199915014207363,
103
+ "p90_m": 0.03714370504021645,
104
+ "p99_m": 0.06730455495417122,
105
+ "mean_elapsed_s": 3.95765625,
106
+ "mean_success_elapsed_s": 2.6996551724137934
107
+ }
108
+ },
109
+ {
110
+ "checkpoint": "model_125.pt",
111
+ "report": "model_125-development.json",
112
+ "summary": {
113
+ "episodes": 128,
114
+ "successes": 16,
115
+ "success_rate": 0.125,
116
+ "mean_m": 0.04146010048498283,
117
+ "median_m": 0.039123570546507835,
118
+ "p90_m": 0.07994573712348937,
119
+ "p99_m": 0.10747653596103192,
120
+ "mean_elapsed_s": 4.81609375,
121
+ "mean_success_elapsed_s": 3.52875
122
+ }
123
+ },
124
+ {
125
+ "checkpoint": "model_100.pt",
126
+ "report": "model_100-development.json",
127
+ "summary": {
128
+ "episodes": 128,
129
+ "successes": 0,
130
+ "success_rate": 0.0,
131
+ "mean_m": 0.06457780898199417,
132
+ "median_m": 0.06327097117900848,
133
+ "p90_m": 0.09948156923055648,
134
+ "p99_m": 0.12729549698531628,
135
+ "mean_elapsed_s": 5.0,
136
+ "mean_success_elapsed_s": null
137
+ }
138
+ },
139
+ {
140
+ "checkpoint": "model_75.pt",
141
+ "report": "model_75-development.json",
142
+ "summary": {
143
+ "episodes": 128,
144
+ "successes": 0,
145
+ "success_rate": 0.0,
146
+ "mean_m": 0.09021183228469454,
147
+ "median_m": 0.07291378825902939,
148
+ "p90_m": 0.1625783249735832,
149
+ "p99_m": 0.23293452143669144,
150
+ "mean_elapsed_s": 5.0,
151
+ "mean_success_elapsed_s": null
152
+ }
153
+ },
154
+ {
155
+ "checkpoint": "model_50.pt",
156
+ "report": "model_50-development.json",
157
+ "summary": {
158
+ "episodes": 128,
159
+ "successes": 0,
160
+ "success_rate": 0.0,
161
+ "mean_m": 0.12645898789924104,
162
+ "median_m": 0.11690208315849304,
163
+ "p90_m": 0.21259460747241973,
164
+ "p99_m": 0.25593534201383594,
165
+ "mean_elapsed_s": 5.0,
166
+ "mean_success_elapsed_s": null
167
+ }
168
+ },
169
+ {
170
+ "checkpoint": "model_25.pt",
171
+ "report": "model_25-development.json",
172
+ "summary": {
173
+ "episodes": 128,
174
+ "successes": 0,
175
+ "success_rate": 0.0,
176
+ "mean_m": 0.2034637692850083,
177
+ "median_m": 0.18847760558128357,
178
+ "p90_m": 0.3301383078098297,
179
+ "p99_m": 0.4213224533200265,
180
+ "mean_elapsed_s": 5.0,
181
+ "mean_success_elapsed_s": null
182
+ }
183
+ },
184
+ {
185
+ "checkpoint": "model_0.pt",
186
+ "report": "model_0-development.json",
187
+ "summary": {
188
+ "episodes": 128,
189
+ "successes": 0,
190
+ "success_rate": 0.0,
191
+ "mean_m": 0.2563139653066173,
192
+ "median_m": 0.26989611983299255,
193
+ "p90_m": 0.36246353983879087,
194
+ "p99_m": 0.406215943992138,
195
+ "mean_elapsed_s": 5.0,
196
+ "mean_success_elapsed_s": null
197
+ }
198
+ }
199
+ ],
200
+ "development_seed": 17002,
201
+ "development_episodes": 128,
202
+ "ranking": [
203
+ "success_rate_desc",
204
+ "mean_m_asc",
205
+ "mean_success_elapsed_s_asc"
206
+ ]
207
+ }
reach/xarm5-reach/seed-7/config.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 1
2
+ task: reach
3
+ robot_key: xarm5_1305
4
+ env:
5
+ num_envs: 4096
6
+ genesis_performance_mode: true
7
+ randomize_initial_episode_length: true
8
+ task_name: reach
9
+ env_version: 2
10
+ num_obs: 21
11
+ num_actions: 5
12
+ runtime_config_sha256: 0471650544c9e38f0bb57a8ce94edc3c4e798fa1961a4028042a2826edc2e7c1
13
+ episode_length_s: 5.0
14
+ ctrl_dt: 0.02
15
+ success_distance_m: 0.005
16
+ action_scale: 0.05
17
+ action_clip: 1.0
18
+ servo_speed_rad_s: 0.5
19
+ max_joint_delta_rad: 0.01
20
+ target_pos_lower:
21
+ - 0.15
22
+ - -0.3
23
+ - 0.05
24
+ target_pos_upper:
25
+ - 0.55
26
+ - 0.3
27
+ - 0.5
28
+ substeps: 32
29
+ constraint_solver: newton
30
+ solver_iterations: 100
31
+ noslip_iterations: 0
32
+ friction_cone: pyramidal
33
+ contact_resolution: convex
34
+ constraint_time_constant_s: 0.00125
35
+ use_gjk_collision: true
36
+ reward:
37
+ reach: 1.0
38
+ action_penalty: 0.001
39
+ success: 2.0
40
+ robot:
41
+ urdf_path: assets/urdf/xarm5/xarm5_1305.urdf
42
+ base_pos:
43
+ - 0.0
44
+ - 0.0
45
+ - 0.0
46
+ ee_link_name: link5
47
+ joint_names:
48
+ - joint1
49
+ - joint2
50
+ - joint3
51
+ - joint4
52
+ - joint5
53
+ default_qpos:
54
+ - 0.0
55
+ - -0.526841
56
+ - -0.402569
57
+ - 0.92941
58
+ - 0.0
59
+ kp:
60
+ - 3000.0
61
+ - 3000.0
62
+ - 2000.0
63
+ - 2000.0
64
+ - 1000.0
65
+ kv:
66
+ - 300.0
67
+ - 300.0
68
+ - 200.0
69
+ - 200.0
70
+ - 100.0
71
+ force_lower:
72
+ - -179.4445
73
+ - -179.4445
74
+ - -92.0
75
+ - -30.6
76
+ - -30.6
77
+ force_upper:
78
+ - 179.4445
79
+ - 179.4445
80
+ - 92.0
81
+ - 30.6
82
+ - 30.6
83
+ train:
84
+ algorithm:
85
+ class_name: PPO
86
+ clip_param: 0.2
87
+ desired_kl: 0.01
88
+ entropy_coef: 0.001
89
+ gamma: 0.99
90
+ lam: 0.95
91
+ learning_rate: 0.0003
92
+ max_grad_norm: 1.0
93
+ num_learning_epochs: 5
94
+ num_mini_batches: 4
95
+ schedule: fixed
96
+ use_clipped_value_loss: true
97
+ value_loss_coef: 1.0
98
+ rnd_cfg: null
99
+ symmetry_cfg: null
100
+ actor:
101
+ class_name: MLPModel
102
+ hidden_dims:
103
+ - 256
104
+ - 256
105
+ - 128
106
+ activation: elu
107
+ obs_normalization: true
108
+ distribution_cfg:
109
+ class_name: BetaDistribution
110
+ action_range:
111
+ - -1.0
112
+ - 1.0
113
+ critic:
114
+ class_name: MLPModel
115
+ hidden_dims:
116
+ - 256
117
+ - 256
118
+ - 128
119
+ activation: elu
120
+ obs_normalization: true
121
+ obs_groups:
122
+ actor:
123
+ - policy
124
+ critic:
125
+ - policy
126
+ runner:
127
+ checkpoint: -1
128
+ experiment_name: xarm5-reach
129
+ load_run: -1
130
+ log_interval: 1
131
+ max_iterations: 300
132
+ record_interval: -1
133
+ resume: false
134
+ resume_path: null
135
+ run_name: ''
136
+ num_steps_per_env: 24
137
+ save_interval: 25
138
+ seed: 7
139
+ run_name: ''
140
+ logger: tensorboard
141
+ config_sha256: 018f085e79e618488b28d7e90f80f805359dd0d41133aed82fef9f13a0943aba
reach/xarm5-reach/seed-7/model_299.checkpoint_manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 5,
3
+ "checkpoint_file": "model_299.pt",
4
+ "checkpoint_sha256": "a53cafcf023b6fab8aecde93f25763878dc76ceebc243cf57579cd4caa7962c1",
5
+ "code_version": "c4aaa345ec0d702ae7801373965c0b5ccf5f7576",
6
+ "config_sha256": "018f085e79e618488b28d7e90f80f805359dd0d41133aed82fef9f13a0943aba",
7
+ "dof": 5,
8
+ "executor_action_contract": "normalized_joint_position_delta",
9
+ "observation_dim": 21,
10
+ "robot_key": "xarm5_1305",
11
+ "schema_version": 1,
12
+ "task": "reach"
13
+ }
reach/xarm5-reach/seed-7/model_299.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a53cafcf023b6fab8aecde93f25763878dc76ceebc243cf57579cd4caa7962c1
3
+ size 2956380
reach/xarm6-reach/seed-7/acceptance-result.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_seed": 7,
3
+ "selected_checkpoint": "model_299.pt",
4
+ "passed": true,
5
+ "holdouts": [
6
+ {
7
+ "seed": 17000,
8
+ "summary": {
9
+ "episodes": 512,
10
+ "successes": 512,
11
+ "success_rate": 1.0,
12
+ "mean_m": 0.0042307584271839005,
13
+ "median_m": 0.004349258961156011,
14
+ "p90_m": 0.0049068403895944366,
15
+ "p99_m": 0.004990258496254682,
16
+ "mean_elapsed_s": 1.568046875,
17
+ "mean_success_elapsed_s": 1.568046875
18
+ },
19
+ "report": "acceptance-17000.json"
20
+ },
21
+ {
22
+ "seed": 17001,
23
+ "summary": {
24
+ "episodes": 512,
25
+ "successes": 512,
26
+ "success_rate": 1.0,
27
+ "mean_m": 0.004240843366460467,
28
+ "median_m": 0.004360834369435906,
29
+ "p90_m": 0.0049156831111758946,
30
+ "p99_m": 0.004987848713062704,
31
+ "mean_elapsed_s": 1.5338671875,
32
+ "mean_success_elapsed_s": 1.5338671875
33
+ },
34
+ "report": "acceptance-17001.json"
35
+ }
36
+ ]
37
+ }
reach/xarm6-reach/seed-7/checkpoint-selection.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "selected_checkpoint": "model_299.pt",
3
+ "candidates": [
4
+ {
5
+ "checkpoint": "model_299.pt",
6
+ "report": "model_299-development.json",
7
+ "summary": {
8
+ "episodes": 128,
9
+ "successes": 128,
10
+ "success_rate": 1.0,
11
+ "mean_m": 0.004202292033369304,
12
+ "median_m": 0.004367312882095575,
13
+ "p90_m": 0.004878329299390316,
14
+ "p99_m": 0.004958784440532327,
15
+ "mean_elapsed_s": 1.5740625,
16
+ "mean_success_elapsed_s": 1.5740625
17
+ }
18
+ },
19
+ {
20
+ "checkpoint": "model_275.pt",
21
+ "report": "model_275-development.json",
22
+ "summary": {
23
+ "episodes": 128,
24
+ "successes": 128,
25
+ "success_rate": 1.0,
26
+ "mean_m": 0.0042429619970789645,
27
+ "median_m": 0.004358453908935189,
28
+ "p90_m": 0.004901017574593425,
29
+ "p99_m": 0.004993231045082211,
30
+ "mean_elapsed_s": 1.59,
31
+ "mean_success_elapsed_s": 1.59
32
+ }
33
+ },
34
+ {
35
+ "checkpoint": "model_225.pt",
36
+ "report": "model_225-development.json",
37
+ "summary": {
38
+ "episodes": 128,
39
+ "successes": 128,
40
+ "success_rate": 1.0,
41
+ "mean_m": 0.004396776621433673,
42
+ "median_m": 0.004505112301558256,
43
+ "p90_m": 0.004898784914985299,
44
+ "p99_m": 0.004990081442520023,
45
+ "mean_elapsed_s": 1.62640625,
46
+ "mean_success_elapsed_s": 1.62640625
47
+ }
48
+ },
49
+ {
50
+ "checkpoint": "model_200.pt",
51
+ "report": "model_200-development.json",
52
+ "summary": {
53
+ "episodes": 128,
54
+ "successes": 128,
55
+ "success_rate": 1.0,
56
+ "mean_m": 0.004397701461130055,
57
+ "median_m": 0.004478892311453819,
58
+ "p90_m": 0.004842542810365558,
59
+ "p99_m": 0.0049754542717710135,
60
+ "mean_elapsed_s": 1.69265625,
61
+ "mean_success_elapsed_s": 1.69265625
62
+ }
63
+ },
64
+ {
65
+ "checkpoint": "model_250.pt",
66
+ "report": "model_250-development.json",
67
+ "summary": {
68
+ "episodes": 128,
69
+ "successes": 127,
70
+ "success_rate": 0.9921875,
71
+ "mean_m": 0.0044319854059722275,
72
+ "median_m": 0.004515522625297308,
73
+ "p90_m": 0.004896547924727201,
74
+ "p99_m": 0.004981113937683404,
75
+ "mean_elapsed_s": 1.630625,
76
+ "mean_success_elapsed_s": 1.6040944881889763
77
+ }
78
+ },
79
+ {
80
+ "checkpoint": "model_175.pt",
81
+ "report": "model_175-development.json",
82
+ "summary": {
83
+ "episodes": 128,
84
+ "successes": 127,
85
+ "success_rate": 0.9921875,
86
+ "mean_m": 0.004560703653623932,
87
+ "median_m": 0.004661683924496174,
88
+ "p90_m": 0.004942381987348199,
89
+ "p99_m": 0.004991176780313253,
90
+ "mean_elapsed_s": 1.7862500000000001,
91
+ "mean_success_elapsed_s": 1.760944881889764
92
+ }
93
+ },
94
+ {
95
+ "checkpoint": "model_150.pt",
96
+ "report": "model_150-development.json",
97
+ "summary": {
98
+ "episodes": 128,
99
+ "successes": 124,
100
+ "success_rate": 0.96875,
101
+ "mean_m": 0.004823907755053369,
102
+ "median_m": 0.004789935424923897,
103
+ "p90_m": 0.004977001948282123,
104
+ "p99_m": 0.00929920053109527,
105
+ "mean_elapsed_s": 2.03984375,
106
+ "mean_success_elapsed_s": 1.9443548387096774
107
+ }
108
+ },
109
+ {
110
+ "checkpoint": "model_125.pt",
111
+ "report": "model_125-development.json",
112
+ "summary": {
113
+ "episodes": 128,
114
+ "successes": 118,
115
+ "success_rate": 0.921875,
116
+ "mean_m": 0.00509433035949769,
117
+ "median_m": 0.004847333766520023,
118
+ "p90_m": 0.004994399240240454,
119
+ "p99_m": 0.014759318893775355,
120
+ "mean_elapsed_s": 2.41984375,
121
+ "mean_success_elapsed_s": 2.201186440677966
122
+ }
123
+ },
124
+ {
125
+ "checkpoint": "model_100.pt",
126
+ "report": "model_100-development.json",
127
+ "summary": {
128
+ "episodes": 128,
129
+ "successes": 79,
130
+ "success_rate": 0.6171875,
131
+ "mean_m": 0.011407616750148009,
132
+ "median_m": 0.004944006679579616,
133
+ "p90_m": 0.027746491134166718,
134
+ "p99_m": 0.04591295816004277,
135
+ "mean_elapsed_s": 3.38140625,
136
+ "mean_success_elapsed_s": 2.3774683544303796
137
+ }
138
+ },
139
+ {
140
+ "checkpoint": "model_75.pt",
141
+ "report": "model_75-development.json",
142
+ "summary": {
143
+ "episodes": 128,
144
+ "successes": 14,
145
+ "success_rate": 0.109375,
146
+ "mean_m": 0.04109923738724319,
147
+ "median_m": 0.029729651287198067,
148
+ "p90_m": 0.08438911139965057,
149
+ "p99_m": 0.11742482110857964,
150
+ "mean_elapsed_s": 4.74546875,
151
+ "mean_success_elapsed_s": 2.672857142857143
152
+ }
153
+ },
154
+ {
155
+ "checkpoint": "model_50.pt",
156
+ "report": "model_50-development.json",
157
+ "summary": {
158
+ "episodes": 128,
159
+ "successes": 2,
160
+ "success_rate": 0.015625,
161
+ "mean_m": 0.11096665077639045,
162
+ "median_m": 0.10870949923992157,
163
+ "p90_m": 0.16305938810110093,
164
+ "p99_m": 0.19393266364932063,
165
+ "mean_elapsed_s": 4.97,
166
+ "mean_success_elapsed_s": 3.08
167
+ }
168
+ },
169
+ {
170
+ "checkpoint": "model_25.pt",
171
+ "report": "model_25-development.json",
172
+ "summary": {
173
+ "episodes": 128,
174
+ "successes": 0,
175
+ "success_rate": 0.0,
176
+ "mean_m": 0.184258812929329,
177
+ "median_m": 0.15868879109621048,
178
+ "p90_m": 0.3252659976482391,
179
+ "p99_m": 0.5026247110962869,
180
+ "mean_elapsed_s": 5.0,
181
+ "mean_success_elapsed_s": null
182
+ }
183
+ },
184
+ {
185
+ "checkpoint": "model_0.pt",
186
+ "report": "model_0-development.json",
187
+ "summary": {
188
+ "episodes": 128,
189
+ "successes": 0,
190
+ "success_rate": 0.0,
191
+ "mean_m": 0.272888166597113,
192
+ "median_m": 0.28472286462783813,
193
+ "p90_m": 0.38783294558525083,
194
+ "p99_m": 0.41890596121549606,
195
+ "mean_elapsed_s": 5.0,
196
+ "mean_success_elapsed_s": null
197
+ }
198
+ }
199
+ ],
200
+ "development_seed": 17002,
201
+ "development_episodes": 128,
202
+ "ranking": [
203
+ "success_rate_desc",
204
+ "mean_m_asc",
205
+ "mean_success_elapsed_s_asc"
206
+ ]
207
+ }
reach/xarm6-reach/seed-7/config.yaml ADDED
@@ -0,0 +1,147 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 1
2
+ task: reach
3
+ robot_key: xarm6_1305
4
+ env:
5
+ num_envs: 4096
6
+ genesis_performance_mode: true
7
+ randomize_initial_episode_length: true
8
+ task_name: reach
9
+ env_version: 2
10
+ num_obs: 24
11
+ num_actions: 6
12
+ runtime_config_sha256: a6a536f4c0e2193a0ccc6bc836521f65c1cf0dea0a493c74caf0d09f74e72d34
13
+ episode_length_s: 5.0
14
+ ctrl_dt: 0.02
15
+ success_distance_m: 0.005
16
+ action_scale: 0.05
17
+ action_clip: 1.0
18
+ servo_speed_rad_s: 0.5
19
+ max_joint_delta_rad: 0.01
20
+ target_pos_lower:
21
+ - 0.15
22
+ - -0.3
23
+ - 0.05
24
+ target_pos_upper:
25
+ - 0.55
26
+ - 0.3
27
+ - 0.5
28
+ substeps: 32
29
+ constraint_solver: newton
30
+ solver_iterations: 100
31
+ noslip_iterations: 0
32
+ friction_cone: pyramidal
33
+ contact_resolution: convex
34
+ constraint_time_constant_s: 0.00125
35
+ use_gjk_collision: true
36
+ reward:
37
+ reach: 1.0
38
+ action_penalty: 0.001
39
+ success: 2.0
40
+ robot:
41
+ urdf_path: assets/urdf/xarm6/xarm6_1305.urdf
42
+ base_pos:
43
+ - 0.0
44
+ - 0.0
45
+ - 0.0
46
+ ee_link_name: link6
47
+ joint_names:
48
+ - joint1
49
+ - joint2
50
+ - joint3
51
+ - joint4
52
+ - joint5
53
+ - joint6
54
+ default_qpos:
55
+ - 0.0
56
+ - -0.526841
57
+ - -0.402569
58
+ - 0.0
59
+ - 0.92941
60
+ - 0.0
61
+ kp:
62
+ - 3000.0
63
+ - 3000.0
64
+ - 2000.0
65
+ - 2000.0
66
+ - 1000.0
67
+ - 1000.0
68
+ kv:
69
+ - 300.0
70
+ - 300.0
71
+ - 200.0
72
+ - 200.0
73
+ - 100.0
74
+ - 100.0
75
+ force_lower:
76
+ - -179.4445
77
+ - -179.4445
78
+ - -92.0
79
+ - -81.558
80
+ - -30.6
81
+ - -30.6
82
+ force_upper:
83
+ - 179.4445
84
+ - 179.4445
85
+ - 92.0
86
+ - 81.558
87
+ - 30.6
88
+ - 30.6
89
+ train:
90
+ algorithm:
91
+ class_name: PPO
92
+ clip_param: 0.2
93
+ desired_kl: 0.01
94
+ entropy_coef: 0.001
95
+ gamma: 0.99
96
+ lam: 0.95
97
+ learning_rate: 0.0003
98
+ max_grad_norm: 1.0
99
+ num_learning_epochs: 5
100
+ num_mini_batches: 4
101
+ schedule: fixed
102
+ use_clipped_value_loss: true
103
+ value_loss_coef: 1.0
104
+ rnd_cfg: null
105
+ symmetry_cfg: null
106
+ actor:
107
+ class_name: MLPModel
108
+ hidden_dims:
109
+ - 256
110
+ - 256
111
+ - 128
112
+ activation: elu
113
+ obs_normalization: true
114
+ distribution_cfg:
115
+ class_name: BetaDistribution
116
+ action_range:
117
+ - -1.0
118
+ - 1.0
119
+ critic:
120
+ class_name: MLPModel
121
+ hidden_dims:
122
+ - 256
123
+ - 256
124
+ - 128
125
+ activation: elu
126
+ obs_normalization: true
127
+ obs_groups:
128
+ actor:
129
+ - policy
130
+ critic:
131
+ - policy
132
+ runner:
133
+ checkpoint: -1
134
+ experiment_name: xarm6-reach
135
+ load_run: -1
136
+ log_interval: 1
137
+ max_iterations: 300
138
+ record_interval: -1
139
+ resume: false
140
+ resume_path: null
141
+ run_name: ''
142
+ num_steps_per_env: 24
143
+ save_interval: 25
144
+ seed: 7
145
+ run_name: ''
146
+ logger: tensorboard
147
+ config_sha256: 1bed49899e3b427dc6cc7b9803f3208140eda0decf07cc0201d5d6d4dd82569e
reach/xarm6-reach/seed-7/model_299.checkpoint_manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 6,
3
+ "checkpoint_file": "model_299.pt",
4
+ "checkpoint_sha256": "6d1b7c16f1a0b0bd58bd3e8d8a3582193d1b584c1fba3f1387fbb22ea5508e73",
5
+ "code_version": "c4aaa345ec0d702ae7801373965c0b5ccf5f7576",
6
+ "config_sha256": "1bed49899e3b427dc6cc7b9803f3208140eda0decf07cc0201d5d6d4dd82569e",
7
+ "dof": 6,
8
+ "executor_action_contract": "normalized_joint_position_delta",
9
+ "observation_dim": 24,
10
+ "robot_key": "xarm6_1305",
11
+ "schema_version": 1,
12
+ "task": "reach"
13
+ }
reach/xarm6-reach/seed-7/model_299.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d1b7c16f1a0b0bd58bd3e8d8a3582193d1b584c1fba3f1387fbb22ea5508e73
3
+ size 3027036
reach/xarm7-reach/seed-7/acceptance-result.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_seed": 7,
3
+ "selected_checkpoint": "model_299.pt",
4
+ "passed": true,
5
+ "holdouts": [
6
+ {
7
+ "seed": 17000,
8
+ "summary": {
9
+ "episodes": 512,
10
+ "successes": 511,
11
+ "success_rate": 0.998046875,
12
+ "mean_m": 0.004387350565593806,
13
+ "median_m": 0.00442216987721622,
14
+ "p90_m": 0.004886777000501752,
15
+ "p99_m": 0.004987741429358721,
16
+ "mean_elapsed_s": 1.409453125,
17
+ "mean_success_elapsed_s": 1.402426614481409
18
+ },
19
+ "report": "acceptance-17000.json"
20
+ },
21
+ {
22
+ "seed": 17001,
23
+ "summary": {
24
+ "episodes": 512,
25
+ "successes": 508,
26
+ "success_rate": 0.9921875,
27
+ "mean_m": 0.004516318834248523,
28
+ "median_m": 0.004480018746107817,
29
+ "p90_m": 0.004922519344836474,
30
+ "p99_m": 0.004997306410223246,
31
+ "mean_elapsed_s": 1.4078125000000001,
32
+ "mean_success_elapsed_s": 1.3795275590551181
33
+ },
34
+ "report": "acceptance-17001.json"
35
+ }
36
+ ]
37
+ }
reach/xarm7-reach/seed-7/checkpoint-selection.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "selected_checkpoint": "model_299.pt",
3
+ "candidates": [
4
+ {
5
+ "checkpoint": "model_299.pt",
6
+ "report": "model_299-development.json",
7
+ "summary": {
8
+ "episodes": 128,
9
+ "successes": 128,
10
+ "success_rate": 1.0,
11
+ "mean_m": 0.004285425395210041,
12
+ "median_m": 0.004437942989170551,
13
+ "p90_m": 0.004886608151718974,
14
+ "p99_m": 0.004988327706232667,
15
+ "mean_elapsed_s": 1.399375,
16
+ "mean_success_elapsed_s": 1.399375
17
+ }
18
+ },
19
+ {
20
+ "checkpoint": "model_275.pt",
21
+ "report": "model_275-development.json",
22
+ "summary": {
23
+ "episodes": 128,
24
+ "successes": 128,
25
+ "success_rate": 1.0,
26
+ "mean_m": 0.0043951461902906885,
27
+ "median_m": 0.004437015624716878,
28
+ "p90_m": 0.004887807043269276,
29
+ "p99_m": 0.004981213347055018,
30
+ "mean_elapsed_s": 1.4284375,
31
+ "mean_success_elapsed_s": 1.4284375
32
+ }
33
+ },
34
+ {
35
+ "checkpoint": "model_250.pt",
36
+ "report": "model_250-development.json",
37
+ "summary": {
38
+ "episodes": 128,
39
+ "successes": 128,
40
+ "success_rate": 1.0,
41
+ "mean_m": 0.004403781467772205,
42
+ "median_m": 0.004479569382965565,
43
+ "p90_m": 0.004919250076636672,
44
+ "p99_m": 0.0049911560490727425,
45
+ "mean_elapsed_s": 1.48171875,
46
+ "mean_success_elapsed_s": 1.48171875
47
+ }
48
+ },
49
+ {
50
+ "checkpoint": "model_200.pt",
51
+ "report": "model_200-development.json",
52
+ "summary": {
53
+ "episodes": 128,
54
+ "successes": 128,
55
+ "success_rate": 1.0,
56
+ "mean_m": 0.004655571709008655,
57
+ "median_m": 0.0047834396827965975,
58
+ "p90_m": 0.004962171847000718,
59
+ "p99_m": 0.004993178131990135,
60
+ "mean_elapsed_s": 1.70484375,
61
+ "mean_success_elapsed_s": 1.70484375
62
+ }
63
+ },
64
+ {
65
+ "checkpoint": "model_225.pt",
66
+ "report": "model_225-development.json",
67
+ "summary": {
68
+ "episodes": 128,
69
+ "successes": 127,
70
+ "success_rate": 0.9921875,
71
+ "mean_m": 0.004638055102986982,
72
+ "median_m": 0.004573862534016371,
73
+ "p90_m": 0.004946766747161746,
74
+ "p99_m": 0.004997995318844915,
75
+ "mean_elapsed_s": 1.55734375,
76
+ "mean_success_elapsed_s": 1.530236220472441
77
+ }
78
+ },
79
+ {
80
+ "checkpoint": "model_175.pt",
81
+ "report": "model_175-development.json",
82
+ "summary": {
83
+ "episodes": 128,
84
+ "successes": 113,
85
+ "success_rate": 0.8828125,
86
+ "mean_m": 0.005448420110042207,
87
+ "median_m": 0.0048510918859392405,
88
+ "p90_m": 0.009151857718825339,
89
+ "p99_m": 0.013932764930650593,
90
+ "mean_elapsed_s": 2.2696875,
91
+ "mean_success_elapsed_s": 1.9072566371681416
92
+ }
93
+ },
94
+ {
95
+ "checkpoint": "model_150.pt",
96
+ "report": "model_150-development.json",
97
+ "summary": {
98
+ "episodes": 128,
99
+ "successes": 86,
100
+ "success_rate": 0.671875,
101
+ "mean_m": 0.008792276210442651,
102
+ "median_m": 0.004880584543570876,
103
+ "p90_m": 0.021139489114284513,
104
+ "p99_m": 0.028596079479902985,
105
+ "mean_elapsed_s": 2.9821875,
106
+ "mean_success_elapsed_s": 1.9967441860465116
107
+ }
108
+ },
109
+ {
110
+ "checkpoint": "model_125.pt",
111
+ "report": "model_125-development.json",
112
+ "summary": {
113
+ "episodes": 128,
114
+ "successes": 69,
115
+ "success_rate": 0.5390625,
116
+ "mean_m": 0.014904505896993214,
117
+ "median_m": 0.004960223566740751,
118
+ "p90_m": 0.034181590750813486,
119
+ "p99_m": 0.05283861279487611,
120
+ "mean_elapsed_s": 3.5640625,
121
+ "mean_success_elapsed_s": 2.3362318840579714
122
+ }
123
+ },
124
+ {
125
+ "checkpoint": "model_100.pt",
126
+ "report": "model_100-development.json",
127
+ "summary": {
128
+ "episodes": 128,
129
+ "successes": 10,
130
+ "success_rate": 0.078125,
131
+ "mean_m": 0.03829743782989681,
132
+ "median_m": 0.03666387498378754,
133
+ "p90_m": 0.07039171159267425,
134
+ "p99_m": 0.11641942299902447,
135
+ "mean_elapsed_s": 4.79984375,
136
+ "mean_success_elapsed_s": 2.438
137
+ }
138
+ },
139
+ {
140
+ "checkpoint": "model_75.pt",
141
+ "report": "model_75-development.json",
142
+ "summary": {
143
+ "episodes": 128,
144
+ "successes": 4,
145
+ "success_rate": 0.03125,
146
+ "mean_m": 0.07713590553976246,
147
+ "median_m": 0.06533772870898247,
148
+ "p90_m": 0.14320702850818634,
149
+ "p99_m": 0.17767264261841775,
150
+ "mean_elapsed_s": 4.924375,
151
+ "mean_success_elapsed_s": 2.58
152
+ }
153
+ },
154
+ {
155
+ "checkpoint": "model_50.pt",
156
+ "report": "model_50-development.json",
157
+ "summary": {
158
+ "episodes": 128,
159
+ "successes": 0,
160
+ "success_rate": 0.0,
161
+ "mean_m": 0.11326514539541677,
162
+ "median_m": 0.10070808604359627,
163
+ "p90_m": 0.19646194577217096,
164
+ "p99_m": 0.34783272802829746,
165
+ "mean_elapsed_s": 5.0,
166
+ "mean_success_elapsed_s": null
167
+ }
168
+ },
169
+ {
170
+ "checkpoint": "model_25.pt",
171
+ "report": "model_25-development.json",
172
+ "summary": {
173
+ "episodes": 128,
174
+ "successes": 0,
175
+ "success_rate": 0.0,
176
+ "mean_m": 0.16838963210466318,
177
+ "median_m": 0.14922721683979034,
178
+ "p90_m": 0.2910965532064438,
179
+ "p99_m": 0.3871403008699417,
180
+ "mean_elapsed_s": 5.0,
181
+ "mean_success_elapsed_s": null
182
+ }
183
+ },
184
+ {
185
+ "checkpoint": "model_0.pt",
186
+ "report": "model_0-development.json",
187
+ "summary": {
188
+ "episodes": 128,
189
+ "successes": 0,
190
+ "success_rate": 0.0,
191
+ "mean_m": 0.24731624993728474,
192
+ "median_m": 0.25476691126823425,
193
+ "p90_m": 0.3548655331134796,
194
+ "p99_m": 0.4016451993584633,
195
+ "mean_elapsed_s": 5.0,
196
+ "mean_success_elapsed_s": null
197
+ }
198
+ }
199
+ ],
200
+ "development_seed": 17002,
201
+ "development_episodes": 128,
202
+ "ranking": [
203
+ "success_rate_desc",
204
+ "mean_m_asc",
205
+ "mean_success_elapsed_s_asc"
206
+ ]
207
+ }
reach/xarm7-reach/seed-7/config.yaml ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 1
2
+ task: reach
3
+ robot_key: xarm7_1305
4
+ env:
5
+ num_envs: 4096
6
+ genesis_performance_mode: true
7
+ randomize_initial_episode_length: true
8
+ task_name: reach
9
+ env_version: 2
10
+ num_obs: 27
11
+ num_actions: 7
12
+ runtime_config_sha256: e033fd94dae3e5351cc00f1b8b66d20987fcc6c32b08fecc03e6514ac3e7fd77
13
+ episode_length_s: 5.0
14
+ ctrl_dt: 0.02
15
+ success_distance_m: 0.005
16
+ action_scale: 0.05
17
+ action_clip: 1.0
18
+ servo_speed_rad_s: 0.5
19
+ max_joint_delta_rad: 0.01
20
+ target_pos_lower:
21
+ - 0.15
22
+ - -0.3
23
+ - 0.05
24
+ target_pos_upper:
25
+ - 0.55
26
+ - 0.3
27
+ - 0.5
28
+ substeps: 32
29
+ constraint_solver: newton
30
+ solver_iterations: 100
31
+ noslip_iterations: 0
32
+ friction_cone: pyramidal
33
+ contact_resolution: convex
34
+ constraint_time_constant_s: 0.00125
35
+ use_gjk_collision: true
36
+ reward:
37
+ reach: 1.0
38
+ action_penalty: 0.001
39
+ success: 2.0
40
+ robot:
41
+ urdf_path: assets/urdf/xarm7/xarm7_1305.urdf
42
+ base_pos:
43
+ - 0.0
44
+ - 0.0
45
+ - 0.0
46
+ ee_link_name: link7
47
+ joint_names:
48
+ - joint1
49
+ - joint2
50
+ - joint3
51
+ - joint4
52
+ - joint5
53
+ - joint6
54
+ - joint7
55
+ default_qpos:
56
+ - 0.0
57
+ - -0.491643
58
+ - 0.0
59
+ - 0.405552
60
+ - 0.0
61
+ - 0.897195
62
+ - 0.0
63
+ kp:
64
+ - 3000.0
65
+ - 3000.0
66
+ - 2000.0
67
+ - 2000.0
68
+ - 1000.0
69
+ - 1000.0
70
+ - 800.0
71
+ kv:
72
+ - 300.0
73
+ - 300.0
74
+ - 200.0
75
+ - 200.0
76
+ - 100.0
77
+ - 100.0
78
+ - 80.0
79
+ force_lower:
80
+ - -179.4445
81
+ - -179.4445
82
+ - -92.0
83
+ - -92.0
84
+ - -81.558
85
+ - -30.6
86
+ - -30.6
87
+ force_upper:
88
+ - 179.4445
89
+ - 179.4445
90
+ - 92.0
91
+ - 92.0
92
+ - 81.558
93
+ - 30.6
94
+ - 30.6
95
+ train:
96
+ algorithm:
97
+ class_name: PPO
98
+ clip_param: 0.2
99
+ desired_kl: 0.01
100
+ entropy_coef: 0.001
101
+ gamma: 0.99
102
+ lam: 0.95
103
+ learning_rate: 0.0003
104
+ max_grad_norm: 1.0
105
+ num_learning_epochs: 5
106
+ num_mini_batches: 4
107
+ schedule: fixed
108
+ use_clipped_value_loss: true
109
+ value_loss_coef: 1.0
110
+ rnd_cfg: null
111
+ symmetry_cfg: null
112
+ actor:
113
+ class_name: MLPModel
114
+ hidden_dims:
115
+ - 256
116
+ - 256
117
+ - 128
118
+ activation: elu
119
+ obs_normalization: true
120
+ distribution_cfg:
121
+ class_name: BetaDistribution
122
+ action_range:
123
+ - -1.0
124
+ - 1.0
125
+ critic:
126
+ class_name: MLPModel
127
+ hidden_dims:
128
+ - 256
129
+ - 256
130
+ - 128
131
+ activation: elu
132
+ obs_normalization: true
133
+ obs_groups:
134
+ actor:
135
+ - policy
136
+ critic:
137
+ - policy
138
+ runner:
139
+ checkpoint: -1
140
+ experiment_name: xarm7-reach
141
+ load_run: -1
142
+ log_interval: 1
143
+ max_iterations: 300
144
+ record_interval: -1
145
+ resume: false
146
+ resume_path: null
147
+ run_name: ''
148
+ num_steps_per_env: 24
149
+ save_interval: 25
150
+ seed: 7
151
+ run_name: ''
152
+ logger: tensorboard
153
+ config_sha256: ba2cbccc0b95d42908c2a6216f19bd22f0c4ee4859d7fcec25959916549cb6c5
reach/xarm7-reach/seed-7/model_299.checkpoint_manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 7,
3
+ "checkpoint_file": "model_299.pt",
4
+ "checkpoint_sha256": "ce7b1e515e944d9656991fdce2ee2d948d177fd88de47113d6f45352279a6f92",
5
+ "code_version": "c4aaa345ec0d702ae7801373965c0b5ccf5f7576",
6
+ "config_sha256": "ba2cbccc0b95d42908c2a6216f19bd22f0c4ee4859d7fcec25959916549cb6c5",
7
+ "dof": 7,
8
+ "executor_action_contract": "normalized_joint_position_delta",
9
+ "observation_dim": 27,
10
+ "robot_key": "xarm7_1305",
11
+ "schema_version": 1,
12
+ "task": "reach"
13
+ }
reach/xarm7-reach/seed-7/model_299.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce7b1e515e944d9656991fdce2ee2d948d177fd88de47113d6f45352279a6f92
3
+ size 3097692