dora-openarm-classifier 0.1.0__tar.gz

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.
@@ -0,0 +1,202 @@
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
@@ -0,0 +1,127 @@
1
+ Metadata-Version: 2.4
2
+ Name: dora-openarm-classifier
3
+ Version: 0.1.0
4
+ Summary: dora-rs node that classifies whether the current state completes the task successfully or not
5
+ Author: Enactic, Inc.
6
+ License: Apache-2.0
7
+ Project-URL: Changelog, https://github.com/enactic/dora-openarm-classifier/releases
8
+ Project-URL: Issues, https://github.com/enactic/dora-openarm-classifier/issues
9
+ Project-URL: Repository, https://github.com/enactic/dora-openarm-classifier
10
+ Requires-Python: >=3.10
11
+ Description-Content-Type: text/markdown
12
+ License-File: LICENSE
13
+ Requires-Dist: accelerate
14
+ Requires-Dist: dora-rs>=0.5.0
15
+ Requires-Dist: fastapi
16
+ Requires-Dist: numpy
17
+ Requires-Dist: pillow
18
+ Requires-Dist: pyarrow
19
+ Requires-Dist: python-multipart
20
+ Requires-Dist: qwen-vl-utils
21
+ Requires-Dist: httpx
22
+ Requires-Dist: torch==2.8.*
23
+ Requires-Dist: torchvision==0.23.*
24
+ Requires-Dist: transformers
25
+ Requires-Dist: uvicorn
26
+ Dynamic: license-file
27
+
28
+ # dora-openarm-classifier
29
+
30
+ dora-rs node that classifies whether the current state completes the
31
+ task successfully or not.
32
+
33
+ ## Install
34
+
35
+ ```bash
36
+ pip install dora-openarm-classifier
37
+ ```
38
+
39
+ ## Usage
40
+
41
+ This package provides two entry points:
42
+
43
+ - `dora-openarm-classifier`: the dora-rs node.
44
+ - `dora-openarm-classifier-server`: an optional standalone HTTP server
45
+ that loads the model once and serves classification requests, so the
46
+ node itself does not have to hold the model in memory.
47
+
48
+ For each incoming camera frame the node runs single-frame
49
+ [TOPReward](https://topreward.github.io/webpage/), keeps a sliding
50
+ window of the last `--window` scores, and reports the window median as
51
+ the verdict. When arm positions are provided, `SUCCESS` is only latched
52
+ once the grippers are open (release) **and** the vision window agrees,
53
+ combining vision and proprioception.
54
+
55
+ Use it in a dora dataflow:
56
+
57
+ ```yaml
58
+ nodes:
59
+ - id: classifier
60
+ build: pip install dora-openarm-classifier
61
+ path: dora-openarm-classifier
62
+ env:
63
+ QUESTION: "Is the object placed in the box?"
64
+ inputs:
65
+ image: camera/image
66
+ position_right: arm-right/state
67
+ position_left: arm-left/state
68
+ outputs:
69
+ - result
70
+ ```
71
+
72
+ By default the node loads Qwen3-VL locally. To offload inference to the
73
+ server instead, start the server and point the node at it:
74
+
75
+ ```bash
76
+ # Terminal 1: start the model server
77
+ dora-openarm-classifier-server --port 8000
78
+
79
+ # Terminal 2 (or via the dataflow): run the node against the server
80
+ dora-openarm-classifier --question "..." --server-url http://127.0.0.1:8000
81
+ ```
82
+
83
+ ## Inputs
84
+
85
+ | Input | Description |
86
+ | ---------------- | ------------------------------------------------------------------------------------------------- |
87
+ | `image` | JPEG bytes (uint8 array) with `width`/`height`/`encoding` metadata. Triggers a classification. |
88
+ | `position` | Optional 16-dim arm state `[right(8), left(8)]`; gripper joints at indices 7 and 15. |
89
+ | `position_right` | Optional 8-dim right arm state; gripper joint at index 7. |
90
+ | `position_left` | Optional 8-dim left arm state; gripper joint at index 7. |
91
+
92
+ Positions may be plain arrays or `StructArray`s with a `qpos` field.
93
+ They are stored and applied to the next incoming `image`. Without any
94
+ position input, the node falls back to a vision-only verdict.
95
+
96
+ ## Outputs
97
+
98
+ | Output | Description |
99
+ | -------- | ----------------------------------------------------------------------------------------------------------- |
100
+ | `result` | Float32 window-median score. Metadata carries `verdict` (`SUCCESS`/`FAIL`) and `frame` (frame counter). |
101
+
102
+ ## Command line options
103
+
104
+ Each option can also be set through the matching environment variable.
105
+
106
+ | Option | Environment variable | Default | Description |
107
+ | --------------------- | -------------------- | ---------------------------- | --------------------------------------------------------------------- |
108
+ | `--question` | `QUESTION` | (required) | Yes/no success question asked about each image. |
109
+ | `--server-url` | `SERVER_URL` | (unset; load model locally) | URL of `dora-openarm-classifier-server`. |
110
+ | `--window` | `WINDOW` | `5` | Sliding-window size in frames. |
111
+ | `--threshold` | `THRESHOLD` | `0.5` | `P_yes` threshold for `SUCCESS`. |
112
+ | `--gripper-threshold` | `GRIPPER_THRESHOLD` | `0.2` | `\|gripper joint\|` above which a gripper counts as open. |
113
+ | `--classify-hz` | `CLASSIFY_HZ` | `0.0` | Max inference rate; frames arriving faster are dropped (`0` = no limit). |
114
+
115
+ The server (`dora-openarm-classifier-server`) accepts `--host`
116
+ (default `127.0.0.1`), `--port` (default `8000`), and `--model-id`
117
+ (default `Qwen/Qwen3-VL-4B-Instruct`).
118
+
119
+ ## License
120
+
121
+ Licensed under the Apache License 2.0. See [LICENSE](LICENSE) for details.
122
+
123
+ Copyright 2026 Enactic, Inc.
124
+
125
+ ## Code of Conduct
126
+
127
+ All participation in the OpenArm project is governed by our [Code of Conduct](CODE_OF_CONDUCT.md).
@@ -0,0 +1,100 @@
1
+ # dora-openarm-classifier
2
+
3
+ dora-rs node that classifies whether the current state completes the
4
+ task successfully or not.
5
+
6
+ ## Install
7
+
8
+ ```bash
9
+ pip install dora-openarm-classifier
10
+ ```
11
+
12
+ ## Usage
13
+
14
+ This package provides two entry points:
15
+
16
+ - `dora-openarm-classifier`: the dora-rs node.
17
+ - `dora-openarm-classifier-server`: an optional standalone HTTP server
18
+ that loads the model once and serves classification requests, so the
19
+ node itself does not have to hold the model in memory.
20
+
21
+ For each incoming camera frame the node runs single-frame
22
+ [TOPReward](https://topreward.github.io/webpage/), keeps a sliding
23
+ window of the last `--window` scores, and reports the window median as
24
+ the verdict. When arm positions are provided, `SUCCESS` is only latched
25
+ once the grippers are open (release) **and** the vision window agrees,
26
+ combining vision and proprioception.
27
+
28
+ Use it in a dora dataflow:
29
+
30
+ ```yaml
31
+ nodes:
32
+ - id: classifier
33
+ build: pip install dora-openarm-classifier
34
+ path: dora-openarm-classifier
35
+ env:
36
+ QUESTION: "Is the object placed in the box?"
37
+ inputs:
38
+ image: camera/image
39
+ position_right: arm-right/state
40
+ position_left: arm-left/state
41
+ outputs:
42
+ - result
43
+ ```
44
+
45
+ By default the node loads Qwen3-VL locally. To offload inference to the
46
+ server instead, start the server and point the node at it:
47
+
48
+ ```bash
49
+ # Terminal 1: start the model server
50
+ dora-openarm-classifier-server --port 8000
51
+
52
+ # Terminal 2 (or via the dataflow): run the node against the server
53
+ dora-openarm-classifier --question "..." --server-url http://127.0.0.1:8000
54
+ ```
55
+
56
+ ## Inputs
57
+
58
+ | Input | Description |
59
+ | ---------------- | ------------------------------------------------------------------------------------------------- |
60
+ | `image` | JPEG bytes (uint8 array) with `width`/`height`/`encoding` metadata. Triggers a classification. |
61
+ | `position` | Optional 16-dim arm state `[right(8), left(8)]`; gripper joints at indices 7 and 15. |
62
+ | `position_right` | Optional 8-dim right arm state; gripper joint at index 7. |
63
+ | `position_left` | Optional 8-dim left arm state; gripper joint at index 7. |
64
+
65
+ Positions may be plain arrays or `StructArray`s with a `qpos` field.
66
+ They are stored and applied to the next incoming `image`. Without any
67
+ position input, the node falls back to a vision-only verdict.
68
+
69
+ ## Outputs
70
+
71
+ | Output | Description |
72
+ | -------- | ----------------------------------------------------------------------------------------------------------- |
73
+ | `result` | Float32 window-median score. Metadata carries `verdict` (`SUCCESS`/`FAIL`) and `frame` (frame counter). |
74
+
75
+ ## Command line options
76
+
77
+ Each option can also be set through the matching environment variable.
78
+
79
+ | Option | Environment variable | Default | Description |
80
+ | --------------------- | -------------------- | ---------------------------- | --------------------------------------------------------------------- |
81
+ | `--question` | `QUESTION` | (required) | Yes/no success question asked about each image. |
82
+ | `--server-url` | `SERVER_URL` | (unset; load model locally) | URL of `dora-openarm-classifier-server`. |
83
+ | `--window` | `WINDOW` | `5` | Sliding-window size in frames. |
84
+ | `--threshold` | `THRESHOLD` | `0.5` | `P_yes` threshold for `SUCCESS`. |
85
+ | `--gripper-threshold` | `GRIPPER_THRESHOLD` | `0.2` | `\|gripper joint\|` above which a gripper counts as open. |
86
+ | `--classify-hz` | `CLASSIFY_HZ` | `0.0` | Max inference rate; frames arriving faster are dropped (`0` = no limit). |
87
+
88
+ The server (`dora-openarm-classifier-server`) accepts `--host`
89
+ (default `127.0.0.1`), `--port` (default `8000`), and `--model-id`
90
+ (default `Qwen/Qwen3-VL-4B-Instruct`).
91
+
92
+ ## License
93
+
94
+ Licensed under the Apache License 2.0. See [LICENSE](LICENSE) for details.
95
+
96
+ Copyright 2026 Enactic, Inc.
97
+
98
+ ## Code of Conduct
99
+
100
+ All participation in the OpenArm project is governed by our [Code of Conduct](CODE_OF_CONDUCT.md).
@@ -0,0 +1,61 @@
1
+ # Copyright 2026 Enactic, Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ [project]
16
+ name = "dora-openarm-classifier"
17
+ version = "0.1.0"
18
+ authors = [{ name = "Enactic, Inc." }]
19
+ description = "dora-rs node that classifies whether the current state completes the task successfully or not"
20
+ license = { text = "Apache-2.0" }
21
+ readme = "README.md"
22
+ requires-python = ">=3.10"
23
+
24
+ dependencies = [
25
+ "accelerate",
26
+ "dora-rs>=0.5.0",
27
+ "fastapi",
28
+ "numpy",
29
+ "pillow",
30
+ "pyarrow",
31
+ "python-multipart",
32
+ "qwen-vl-utils",
33
+ "httpx",
34
+ "torch==2.8.*",
35
+ "torchvision==0.23.*",
36
+ "transformers",
37
+ "uvicorn",
38
+ ]
39
+
40
+ [dependency-groups]
41
+ dev = [
42
+ "ruff == 0.14.11", # sync with .pre-commit-config.yaml
43
+ ]
44
+
45
+ [project.scripts]
46
+ dora-openarm-classifier = "dora_openarm_classifier.main:main"
47
+ dora-openarm-classifier-server = "dora_openarm_classifier.server:main"
48
+
49
+ [project.urls]
50
+ Changelog = "https://github.com/enactic/dora-openarm-classifier/releases"
51
+ Issues = "https://github.com/enactic/dora-openarm-classifier/issues"
52
+ Repository = "https://github.com/enactic/dora-openarm-classifier"
53
+
54
+ [tool.ruff.lint]
55
+ extend-select = [
56
+ "D", # pydocstyle
57
+ "UP"
58
+ ]
59
+
60
+ [tool.ruff.lint.per-file-ignores]
61
+ "!src/**.py" = ["D"]
@@ -0,0 +1,4 @@
1
+ [egg_info]
2
+ tag_build =
3
+ tag_date = 0
4
+
@@ -0,0 +1,15 @@
1
+ # Copyright 2026 Enactic, Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ """dora-rs node that classifies whether the current state completes the task successfully or not."""
@@ -0,0 +1,228 @@
1
+ # Copyright 2026 Enactic, Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ """dora-rs node that classifies whether the current state completes the task successfully or not.
16
+
17
+ For each incoming camera frame it runs single-frame TOPReward, keeps a
18
+ sliding window of the last WINDOW scores, and reports the window
19
+ median as the verdict. When arm positions are provided, success is
20
+ only latched once the grippers are open (release) AND the vision
21
+ window agrees -- combining vision and proprioception, like the
22
+ production classifier node.
23
+
24
+ Inputs:
25
+ image JPEG bytes (uint8 array) with width/height/encoding metadata
26
+ position optional 16-dim arm state [right(8), left(8)]; grippers at 7 and 15
27
+ position_right optional 8-dim right arm state; gripper at 7
28
+ position_left optional 8-dim left arm state; gripper at 7
29
+ Outputs:
30
+ result Float32 score; metadata has "verdict" (SUCCESS/FAIL) and "frame"
31
+
32
+ Command line options/Environment variables:
33
+ --question, QUESTION yes/no success question
34
+ --server-url, SERVER_URL URL for dora-openarm-classifier-server
35
+ --window, WINDOW sliding-window size in frames (default 5)
36
+ --threshold, THRESHOLD P_yes threshold for SUCCESS (default 0.5)
37
+ --gripper-threshold, GRIPPER_THRESHOLD
38
+ |gripper joint| above which a gripper counts
39
+ as open (default 0.2)
40
+ --classify-hz, CLASSIFY_HZ max inference rate; drop frames arriving
41
+ faster (0 = no limit)
42
+ """
43
+
44
+ import argparse
45
+ from collections import deque
46
+ import dora
47
+ from PIL import Image
48
+ import httpx
49
+ import io
50
+ import numpy as np
51
+ import os
52
+ import pyarrow as pa
53
+ import time
54
+
55
+ from .topreward import TOPRewardScorer
56
+
57
+
58
+ def _score_remote(args, image_bytes):
59
+ response = httpx.post(
60
+ f"{args.server_url}/classify",
61
+ data={"question": args.question, "threshold": args.threshold},
62
+ files={"image": ("image.jpeg", image_bytes, "image/jpeg")},
63
+ timeout=10,
64
+ )
65
+
66
+ if response.status_code != 200:
67
+ print(
68
+ f"ERROR: Server returned {response.status_code}: {response.content}",
69
+ flush=True,
70
+ )
71
+ return None
72
+
73
+ return response.json()["score"]
74
+
75
+
76
+ def main():
77
+ """Run the real-time TOPReword based success classifier."""
78
+ parser = argparse.ArgumentParser(
79
+ description="Classify whether the current task is completed successully or not"
80
+ )
81
+ default_question = os.getenv("QUESTION")
82
+ parser.add_argument(
83
+ "--question",
84
+ required=default_question is None,
85
+ default=default_question,
86
+ help="Yes/no question to ask about images",
87
+ )
88
+ parser.add_argument(
89
+ "--server-url",
90
+ default=os.getenv("SERVER_URL"),
91
+ help="dora-openarm-classifier-server URL",
92
+ )
93
+ parser.add_argument(
94
+ "--window",
95
+ default=int(os.getenv("WINDOW", "5")),
96
+ help="The number of images to use for classification",
97
+ type=int,
98
+ )
99
+ parser.add_argument(
100
+ "--threshold",
101
+ default=float(os.getenv("THRESHOLD", "0.5")),
102
+ help="P_yes threshold for SUCCESS (default: 0.5)",
103
+ type=float,
104
+ )
105
+ parser.add_argument(
106
+ "--gripper-threshold",
107
+ type=float,
108
+ default=float(os.getenv("GRIPPER_THRESHOLD", "0.2")),
109
+ help="Above which a gripper counts as open (default 0.2)",
110
+ )
111
+ parser.add_argument(
112
+ "--classify-hz",
113
+ type=float,
114
+ default=float(os.getenv("CLASSIFY_HZ", "0.0")),
115
+ # Real cameras run at ~30fps but Qwen is ~2Hz, so cap it and
116
+ # drop frames that arrive too soon.
117
+ help="Max classify rate; drop frames arriving faster (0 = no limit)",
118
+ )
119
+ args = parser.parse_args()
120
+ min_interval = 1.0 / args.classify_hz if args.classify_hz > 0 else 0.0
121
+ last_classify = 0.0
122
+
123
+ if args.server_url:
124
+ print(f"Connecting to server: {args.server_url}", flush=True)
125
+ try:
126
+ httpx.get(f"{args.server_url}/health", timeout=5)
127
+ except Exception as error:
128
+ print(f"Cannot connect to server: {args.server_url}: {error}", flush=True)
129
+ print(
130
+ "Make sure server is running: dora-openarm-classifier-server",
131
+ flush=True,
132
+ )
133
+ raise
134
+ else:
135
+ print("Loading Qwen3-VL ...", flush=True)
136
+ scorer = TOPRewardScorer()
137
+
138
+ print("Ready:", flush=True)
139
+ print(f" window: {args.window}", flush=True)
140
+ print(f" threshold: {args.threshold:.0%}", flush=True)
141
+ print(f" question: {args.question}", flush=True)
142
+
143
+ scores = deque(maxlen=args.window)
144
+ last_right_pos = None # latest right arm position (8-dim)
145
+ last_left_pos = None # latest left arm position (8-dim)
146
+ success_latched = False # sticky once vision + grippers agree
147
+ node = dora.Node()
148
+ n = 0
149
+ for event in node:
150
+ if event["type"] != "INPUT":
151
+ continue
152
+
153
+ if event["id"] == "position":
154
+ value = event["value"]
155
+ if isinstance(value, pa.StructArray):
156
+ value = value.field("qpos")
157
+ last_pos = value.to_numpy(zero_copy_only=False)
158
+ last_right_pos = last_pos[:8]
159
+ last_left_pos = last_pos[8:]
160
+ continue
161
+
162
+ if event["id"] == "position_right":
163
+ value = event["value"]
164
+ if isinstance(value, pa.StructArray):
165
+ value = value.field("qpos")
166
+ last_right_pos = value.to_numpy(zero_copy_only=False)
167
+ continue
168
+
169
+ if event["id"] == "position_left":
170
+ value = event["value"]
171
+ if isinstance(value, pa.StructArray):
172
+ value = value.field("qpos")
173
+ last_left_pos = value.to_numpy(zero_copy_only=False)
174
+ continue
175
+
176
+ if event["id"] == "image":
177
+ # rate-limit: drop frames that arrive faster than INFER_HZ
178
+ now = time.monotonic()
179
+ if min_interval and (now - last_classify) < min_interval:
180
+ continue
181
+ last_classify = now
182
+
183
+ image_bytes = event["value"].to_numpy(zero_copy_only=False).tobytes()
184
+ if args.server_url:
185
+ score = _score_remote(args, image_bytes)
186
+ if score is None:
187
+ continue
188
+ else:
189
+ image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
190
+ _, _, score = scorer.score(image, args.question)
191
+ scores.append(score)
192
+ agg = float(np.median(scores))
193
+
194
+ grip_info = ""
195
+ if (
196
+ last_right_pos is not None
197
+ and len(last_right_pos) >= 8
198
+ and last_left_pos is not None
199
+ and len(last_left_pos) >= 8
200
+ ):
201
+ gr, gl = float(last_right_pos[7]), float(last_left_pos[7])
202
+ grippers_open = (
203
+ abs(gr) > args.gripper_threshold
204
+ and abs(gl) > args.gripper_threshold
205
+ )
206
+ if not success_latched and grippers_open and agg >= args.threshold:
207
+ success_latched = True
208
+ verdict = "SUCCESS" if success_latched else "FAIL"
209
+ grip_info = f" grip(r={gr:+.2f} l={gl:+.2f} {'OPEN' if grippers_open else 'closed'})"
210
+ else:
211
+ # no arm data -> vision-only fallback
212
+ verdict = "SUCCESS" if agg >= args.threshold else "FAIL"
213
+
214
+ n += 1
215
+ print(
216
+ f"frame {n:03d} P_yes={score:5.1%} "
217
+ f"window_median={agg:5.1%}{grip_info} -> {verdict}",
218
+ flush=True,
219
+ )
220
+ node.send_output(
221
+ "result",
222
+ pa.array([agg], type=pa.float32()),
223
+ metadata={"verdict": verdict, "frame": n},
224
+ )
225
+
226
+
227
+ if __name__ == "__main__":
228
+ main()
@@ -0,0 +1,104 @@
1
+ # Copyright 2026 Enactic, Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ """Classification server.
16
+
17
+ Standalone server that loads model once and serves classification requests.
18
+ Clients can send images via HTTP POST and get back success scores.
19
+
20
+ Usage:
21
+ dora-openarm-classifier-server --port 8000
22
+
23
+ API:
24
+ POST /classify
25
+ Content-Type: multipart/form-data
26
+ question: yes/no success question
27
+ image: JPEG or PNG image
28
+ Content-Type: image/jpeg or image/png
29
+
30
+ Response (JSON):
31
+ {
32
+ "p_yes": 0.85,
33
+ "p_no": 0.12,
34
+ "score": 0.876, # normalized p_yes
35
+ }
36
+ """
37
+
38
+ import argparse
39
+ from fastapi import FastAPI, Form, File, UploadFile
40
+ from fastapi.responses import JSONResponse
41
+ from PIL import Image
42
+ from typing import Annotated
43
+ import uvicorn
44
+
45
+ from .topreward import TOPRewardScorer
46
+
47
+ app = FastAPI()
48
+
49
+ scorer = None
50
+ args = None
51
+
52
+
53
+ @app.get("/health", response_class=JSONResponse)
54
+ def _health():
55
+ return {"status": "healthy"}
56
+
57
+
58
+ @app.post("/classify", response_class=JSONResponse)
59
+ async def _classify(
60
+ question: Annotated[str, Form()], image: Annotated[UploadFile, File()]
61
+ ):
62
+ p_yes, p_no, score = scorer.score(Image.open(image.file).convert("RGB"), question)
63
+ return {
64
+ "p_yes": float(p_yes),
65
+ "p_no": float(p_no),
66
+ "score": float(score),
67
+ }
68
+
69
+
70
+ def main():
71
+ """Run TOPReward based classification server."""
72
+ parser = argparse.ArgumentParser(description="Classification server")
73
+ parser.add_argument(
74
+ "--port", type=int, default=8000, help="Server port (default: 8000)"
75
+ )
76
+ parser.add_argument(
77
+ "--host", type=str, default="127.0.0.1", help="Server host (default: 127.0.0.1)"
78
+ )
79
+ parser.add_argument(
80
+ "--model-id",
81
+ type=str,
82
+ default=None,
83
+ help="Model ID (default: Qwen/Qwen3-VL-4B-Instruct)",
84
+ )
85
+
86
+ global args
87
+ args = parser.parse_args()
88
+
89
+ print("Loading model...", flush=True)
90
+ global scorer
91
+ if args.model_id:
92
+ scorer = TOPRewardScorer(args.model_id)
93
+ else:
94
+ scorer = TOPRewardScorer()
95
+ print("Ready!", flush=True)
96
+
97
+ config = uvicorn.Config(app, host=args.host, port=args.port, log_level="info")
98
+ server = uvicorn.Server(config)
99
+
100
+ server.run()
101
+
102
+
103
+ if __name__ == "__main__":
104
+ main()
@@ -0,0 +1,107 @@
1
+ # Copyright 2026 Enactic, Inc.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ """Qwen-VL TOPReward scorer.
16
+
17
+ https://topreward.github.io/webpage/
18
+
19
+ Asks a yes/no question about an image and reads the probability the model
20
+ assigns to "Yes" from the next-token logits. The normalized
21
+ P_yes = P(yes) / (P(yes) + P(no)) is the success score -- no labels, no training.
22
+ """
23
+
24
+ from qwen_vl_utils import process_vision_info
25
+ import torch
26
+ from transformers import AutoProcessor
27
+
28
+ MODEL_ID = "Qwen/Qwen3-VL-4B-Instruct"
29
+ YES_WORDS = ["Yes", "yes", "YES", " Yes", " yes"]
30
+ NO_WORDS = ["No", "no", "NO", " No", " no"]
31
+
32
+
33
+ def _load_model(model_id):
34
+ """Pick the right model class from the id.
35
+
36
+ Use Instruct (not Thinking)
37
+ variants: we read the FIRST generated token, but Thinking models emit
38
+ reasoning first, so the first token would not be Yes/No.
39
+ """
40
+ low = model_id.lower()
41
+ if "qwen3-vl" in low and "moe" in low:
42
+ from transformers import Qwen3VLMoeForConditionalGeneration as Cls
43
+ elif "qwen3-vl" in low:
44
+ from transformers import Qwen3VLForConditionalGeneration as Cls
45
+ else: # qwen2.5-vl and friends
46
+ from transformers import Qwen2_5_VLForConditionalGeneration as Cls
47
+ return Cls.from_pretrained(model_id, torch_dtype="auto", device_map="auto").eval()
48
+
49
+
50
+ class TOPRewardScorer:
51
+ """Single-frame Qwen-VL success scorer."""
52
+
53
+ def __init__(self, model_id=MODEL_ID):
54
+ """Load the model and processor once."""
55
+ print(f"[load] {model_id} ...", flush=True)
56
+ self.model = _load_model(model_id)
57
+ self.proc = AutoProcessor.from_pretrained(model_id)
58
+ self.tok = self.proc.tokenizer
59
+ print(
60
+ f"[load] done. device={self.model.device} dtype={self.model.dtype}\n",
61
+ flush=True,
62
+ )
63
+
64
+ def _mass(self, probs, words):
65
+ s = 0.0
66
+ for w in words:
67
+ ids = self.tok.encode(w, add_special_tokens=False)
68
+ if len(ids) == 1:
69
+ s += float(probs[ids[0]].item())
70
+ return s
71
+
72
+ def next_token_probs(self, image, question):
73
+ """Next-token probability distribution for (image, question).
74
+
75
+ ``image`` may be a file path or a PIL image.
76
+ """
77
+ messages = [
78
+ {
79
+ "role": "user",
80
+ "content": [
81
+ {"type": "image", "image": image},
82
+ {"type": "text", "text": question},
83
+ ],
84
+ }
85
+ ]
86
+ text = self.proc.apply_chat_template(
87
+ messages, tokenize=False, add_generation_prompt=True
88
+ )
89
+ image_inputs, video_inputs = process_vision_info(messages)
90
+ inputs = self.proc(
91
+ text=[text],
92
+ images=image_inputs,
93
+ videos=video_inputs,
94
+ padding=True,
95
+ return_tensors="pt",
96
+ ).to(self.model.device)
97
+ with torch.no_grad():
98
+ logits = self.model(**inputs).logits[0, -1]
99
+ return torch.softmax(logits.float(), dim=-1)
100
+
101
+ def score(self, image, question):
102
+ """Return (raw P_yes, raw P_no, normalized P_yes = TOPReward)."""
103
+ probs = self.next_token_probs(image, question + "\n\nAnswer Yes or No.")
104
+ p_yes = self._mass(probs, YES_WORDS)
105
+ p_no = self._mass(probs, NO_WORDS)
106
+ norm = p_yes / (p_yes + p_no) if (p_yes + p_no) > 0 else 0.0
107
+ return p_yes, p_no, norm
@@ -0,0 +1,127 @@
1
+ Metadata-Version: 2.4
2
+ Name: dora-openarm-classifier
3
+ Version: 0.1.0
4
+ Summary: dora-rs node that classifies whether the current state completes the task successfully or not
5
+ Author: Enactic, Inc.
6
+ License: Apache-2.0
7
+ Project-URL: Changelog, https://github.com/enactic/dora-openarm-classifier/releases
8
+ Project-URL: Issues, https://github.com/enactic/dora-openarm-classifier/issues
9
+ Project-URL: Repository, https://github.com/enactic/dora-openarm-classifier
10
+ Requires-Python: >=3.10
11
+ Description-Content-Type: text/markdown
12
+ License-File: LICENSE
13
+ Requires-Dist: accelerate
14
+ Requires-Dist: dora-rs>=0.5.0
15
+ Requires-Dist: fastapi
16
+ Requires-Dist: numpy
17
+ Requires-Dist: pillow
18
+ Requires-Dist: pyarrow
19
+ Requires-Dist: python-multipart
20
+ Requires-Dist: qwen-vl-utils
21
+ Requires-Dist: httpx
22
+ Requires-Dist: torch==2.8.*
23
+ Requires-Dist: torchvision==0.23.*
24
+ Requires-Dist: transformers
25
+ Requires-Dist: uvicorn
26
+ Dynamic: license-file
27
+
28
+ # dora-openarm-classifier
29
+
30
+ dora-rs node that classifies whether the current state completes the
31
+ task successfully or not.
32
+
33
+ ## Install
34
+
35
+ ```bash
36
+ pip install dora-openarm-classifier
37
+ ```
38
+
39
+ ## Usage
40
+
41
+ This package provides two entry points:
42
+
43
+ - `dora-openarm-classifier`: the dora-rs node.
44
+ - `dora-openarm-classifier-server`: an optional standalone HTTP server
45
+ that loads the model once and serves classification requests, so the
46
+ node itself does not have to hold the model in memory.
47
+
48
+ For each incoming camera frame the node runs single-frame
49
+ [TOPReward](https://topreward.github.io/webpage/), keeps a sliding
50
+ window of the last `--window` scores, and reports the window median as
51
+ the verdict. When arm positions are provided, `SUCCESS` is only latched
52
+ once the grippers are open (release) **and** the vision window agrees,
53
+ combining vision and proprioception.
54
+
55
+ Use it in a dora dataflow:
56
+
57
+ ```yaml
58
+ nodes:
59
+ - id: classifier
60
+ build: pip install dora-openarm-classifier
61
+ path: dora-openarm-classifier
62
+ env:
63
+ QUESTION: "Is the object placed in the box?"
64
+ inputs:
65
+ image: camera/image
66
+ position_right: arm-right/state
67
+ position_left: arm-left/state
68
+ outputs:
69
+ - result
70
+ ```
71
+
72
+ By default the node loads Qwen3-VL locally. To offload inference to the
73
+ server instead, start the server and point the node at it:
74
+
75
+ ```bash
76
+ # Terminal 1: start the model server
77
+ dora-openarm-classifier-server --port 8000
78
+
79
+ # Terminal 2 (or via the dataflow): run the node against the server
80
+ dora-openarm-classifier --question "..." --server-url http://127.0.0.1:8000
81
+ ```
82
+
83
+ ## Inputs
84
+
85
+ | Input | Description |
86
+ | ---------------- | ------------------------------------------------------------------------------------------------- |
87
+ | `image` | JPEG bytes (uint8 array) with `width`/`height`/`encoding` metadata. Triggers a classification. |
88
+ | `position` | Optional 16-dim arm state `[right(8), left(8)]`; gripper joints at indices 7 and 15. |
89
+ | `position_right` | Optional 8-dim right arm state; gripper joint at index 7. |
90
+ | `position_left` | Optional 8-dim left arm state; gripper joint at index 7. |
91
+
92
+ Positions may be plain arrays or `StructArray`s with a `qpos` field.
93
+ They are stored and applied to the next incoming `image`. Without any
94
+ position input, the node falls back to a vision-only verdict.
95
+
96
+ ## Outputs
97
+
98
+ | Output | Description |
99
+ | -------- | ----------------------------------------------------------------------------------------------------------- |
100
+ | `result` | Float32 window-median score. Metadata carries `verdict` (`SUCCESS`/`FAIL`) and `frame` (frame counter). |
101
+
102
+ ## Command line options
103
+
104
+ Each option can also be set through the matching environment variable.
105
+
106
+ | Option | Environment variable | Default | Description |
107
+ | --------------------- | -------------------- | ---------------------------- | --------------------------------------------------------------------- |
108
+ | `--question` | `QUESTION` | (required) | Yes/no success question asked about each image. |
109
+ | `--server-url` | `SERVER_URL` | (unset; load model locally) | URL of `dora-openarm-classifier-server`. |
110
+ | `--window` | `WINDOW` | `5` | Sliding-window size in frames. |
111
+ | `--threshold` | `THRESHOLD` | `0.5` | `P_yes` threshold for `SUCCESS`. |
112
+ | `--gripper-threshold` | `GRIPPER_THRESHOLD` | `0.2` | `\|gripper joint\|` above which a gripper counts as open. |
113
+ | `--classify-hz` | `CLASSIFY_HZ` | `0.0` | Max inference rate; frames arriving faster are dropped (`0` = no limit). |
114
+
115
+ The server (`dora-openarm-classifier-server`) accepts `--host`
116
+ (default `127.0.0.1`), `--port` (default `8000`), and `--model-id`
117
+ (default `Qwen/Qwen3-VL-4B-Instruct`).
118
+
119
+ ## License
120
+
121
+ Licensed under the Apache License 2.0. See [LICENSE](LICENSE) for details.
122
+
123
+ Copyright 2026 Enactic, Inc.
124
+
125
+ ## Code of Conduct
126
+
127
+ All participation in the OpenArm project is governed by our [Code of Conduct](CODE_OF_CONDUCT.md).
@@ -0,0 +1,13 @@
1
+ LICENSE
2
+ README.md
3
+ pyproject.toml
4
+ src/dora_openarm_classifier/__init__.py
5
+ src/dora_openarm_classifier/main.py
6
+ src/dora_openarm_classifier/server.py
7
+ src/dora_openarm_classifier/topreward.py
8
+ src/dora_openarm_classifier.egg-info/PKG-INFO
9
+ src/dora_openarm_classifier.egg-info/SOURCES.txt
10
+ src/dora_openarm_classifier.egg-info/dependency_links.txt
11
+ src/dora_openarm_classifier.egg-info/entry_points.txt
12
+ src/dora_openarm_classifier.egg-info/requires.txt
13
+ src/dora_openarm_classifier.egg-info/top_level.txt
@@ -0,0 +1,3 @@
1
+ [console_scripts]
2
+ dora-openarm-classifier = dora_openarm_classifier.main:main
3
+ dora-openarm-classifier-server = dora_openarm_classifier.server:main
@@ -0,0 +1,13 @@
1
+ accelerate
2
+ dora-rs>=0.5.0
3
+ fastapi
4
+ numpy
5
+ pillow
6
+ pyarrow
7
+ python-multipart
8
+ qwen-vl-utils
9
+ httpx
10
+ torch==2.8.*
11
+ torchvision==0.23.*
12
+ transformers
13
+ uvicorn