dora-openarm-classifier

dora-rs node that classifies whether the current state completes the task successfully or not.

Install

pip install dora-openarm-classifier

Usage

This package provides two entry points:

• dora-openarm-classifier: the dora-rs node.
• dora-openarm-classifier-server: an optional standalone HTTP server that loads the model once and serves classification requests, so the node itself does not have to hold the model in memory.

For each incoming camera frame the node runs single-frame TOPReward, keeps a sliding window of the last --window scores, and reports the window median as the verdict. When arm positions are provided, SUCCESS is only latched once the grippers are open (release) and the vision window agrees, combining vision and proprioception.

Use it in a dora dataflow:

nodes:
  - id: classifier
    build: pip install dora-openarm-classifier
    path: dora-openarm-classifier
    env:
      QUESTION: "Is the object placed in the box?"
    inputs:
      image: camera/image
      position_right: arm-right/state
      position_left: arm-left/state
    outputs:
      - result

By default the node loads Qwen3-VL locally. To offload inference to the server instead, start the server and point the node at it:

# Terminal 1: start the model server
dora-openarm-classifier-server --port 8000

# Terminal 2 (or via the dataflow): run the node against the server
dora-openarm-classifier --question "..." --server-url http://127.0.0.1:8000

Inputs

Positions may be plain arrays or StructArrays with a qpos field. They are stored and applied to the next incoming image. Without any position input, the node falls back to a vision-only verdict.

Outputs

Command line options

Each option can also be set through the matching environment variable.

The server (dora-openarm-classifier-server) accepts --host (default 127.0.0.1), --port (default 8000), and --model-id (default Qwen/Qwen3-VL-4B-Instruct).

License

Licensed under the Apache License 2.0. See LICENSE for details.

Copyright 2026 Enactic, Inc.

Code of Conduct

All participation in the OpenArm project is governed by our Code of Conduct.