Task: OpenArm Integration with DimOS

[mustafab0]

Problem

OpenArm is a bimanual 7-DOF research arm built from Damiao DM-J motors. Unlike every other arm in dimos, it ships no Python SDK — only raw CAN. Need a from-scratch driver + adapter + blueprints to run it.

Closes DIM-386

Solution

From-scratch Damiao MIT-mode CAN driver (ported from enactic/openarm_can) + ManipulatorAdapter wrapper with Pinocchio gravity-comp feedforward + bimanual blueprints. Auto-writes CTRL_MODE=MIT on connect. Per-side URDFs avoid phantom-arm Drake collisions. Generic reachability tool added to dimos.utils since it works on any URDF.

Added:

• dimos/hardware/manipulators/openarm/{driver,adapter,test_driver}.py
• dimos/robot/catalog/openarm.py
• dimos/robot/manipulators/openarm/blueprints.py
• dimos/robot/manipulators/openarm/scripts/{openarm_can_up.sh, openarm_can_probe.py, openarm_set_mit_mode.py}
• dimos/utils/workspace.py
• docs/capabilities/manipulation/openarm_integration.md

Modified: dimos/robot/all_blueprints.py

Breaking Changes

None.

How to Test

Unit tests (no hardware):

.venv/bin/python -m pytest dimos/hardware/manipulators/openarm/ -v

Mock (no hardware, Drake viz):

dimos run openarm-mock-planner-coordinator

Real hardware:

sudo ./dimos/robot/manipulators/openarm/scripts/openarm_can_up.sh can0 can1
python ./dimos/robot/manipulators/openarm/scripts/openarm_can_probe.py --channel can0     # expect 8/8
dimos run openarm-planner-coordinator
# then in another terminal: python -i -m dimos.manipulation.planning.examples.manipulation_client

Full guide: openarm_integration.md.

[greptile-apps]

Greptile Summary

This PR introduces a complete from-scratch OpenArm v10 integration: a Damiao MIT-mode CAN driver (driver.py), a ManipulatorAdapter with Pinocchio gravity-compensation feedforward (adapter.py), bimanual and single-arm blueprints, hardware bring-up scripts, and a generic workspace reachability tool. All previously flagged issues (velocity-mode gravity feedforward, ENOBUFS errno detection, --classical CLI flag) have been addressed. The remaining findings are P2 quality issues only.

Confidence Score: 5/5

Safe to merge — no P0 or P1 issues; all findings are minor style or robustness P2s.

No blocking issues found. The driver, adapter, and tests are well-structured. Previous P1 comments have been addressed. Remaining findings (duplicate import, weak bounds check in a rarely-triggered branch, unclosed file handles in CLI scripts, potential NaN in workspace sampling for unlimited joints) are all P2 and do not affect correctness on the target hardware.

dimos/perception/detection/type/imageDetections.py (duplicate import), dimos/utils/workspace.py (unbounded joint sampling), dimos/robot/manipulators/openarm/scripts/openarm_set_mit_mode.py (struct bounds check)

Important Files Changed

Filename	Overview
dimos/hardware/manipulators/openarm/driver.py	New Damiao MIT-mode CAN driver: frame packing/unpacking, background RX thread with state cache, ENOBUFS retry now using errno directly — clean and well-structured.
dimos/hardware/manipulators/openarm/adapter.py	ManipulatorAdapter wrapping the CAN driver; gravity feedforward now applied in all three write paths (position, velocity, stop) — no issues found.
dimos/hardware/manipulators/openarm/test_driver.py	Good unit-test coverage via virtual CAN loopback: frame packing roundtrips, RX thread state cache, enable/disable, send_mit_many fan-out — all hardware-free.
dimos/perception/detection/type/imageDetections.py	Duplicate Self import block added — appears to be a merge artifact; harmless at runtime but should be removed.
dimos/utils/workspace.py	New reachability + manipulability workspace tool; sampling loop is correct for bounded joints but will silently produce NaN for infinite-limit joints (e.g. continuous revolute or free-floating base).
dimos/robot/manipulators/openarm/scripts/openarm_set_mit_mode.py	Hardware bring-up script; bounds check before struct.unpack is too weak (>=4 instead of >=8) and file handle is unclosed.
dimos/robot/manipulators/openarm/scripts/openarm_can_probe.py	Hardware probe script; uses open() without context manager for sysfs flags read — minor resource leak identical to the pattern in set_mit_mode.py.
dimos/robot/catalog/openarm.py	Clean catalog definitions for single-arm and bimanual configs with proper adapter_kwargs merging; no issues found.
dimos/robot/manipulators/openarm/blueprints.py	Bimanual and single-arm blueprints covering mock, left-only, right-only, and planner+coordinator combos; CAN channel assignment clearly documented.
dimos/robot/all_blueprints.py	Adds eight new OpenArm blueprint entries in alphabetical order — clean registration.

Sequence Diagram

sequenceDiagram
    participant App as Application / Blueprint
    participant Adapter as OpenArmAdapter
    participant Bus as OpenArmBus
    participant RX as RX Thread (daemon)
    participant HW as Damiao Motors (CAN)

    App->>Adapter: connect()
    Adapter->>Bus: open() → start RX thread
    Bus->>RX: spawn thread
    Adapter->>Bus: write_ctrl_mode(MIT) × 7
    Bus->>HW: 0x7FF param-write frame
    Adapter->>Bus: enable_all()
    Bus->>HW: ENABLE frame × 7
    HW-->>RX: state reply frames
    RX->>Bus: _states[recv_id] = MotorState
    Adapter->>Bus: wait_all_states(0.5s)
    Adapter->>Adapter: load Pinocchio model (gravity comp)

    loop Control tick
        App->>Adapter: write_joint_positions(q)
        Adapter->>Bus: get_states() [read lock]
        Adapter->>Adapter: _compute_gravity_torques(q_current)
        Adapter->>Bus: send_mit_many(commands)
        Bus->>HW: MIT frame × 7 (inter-frame 0.5ms)
        HW-->>RX: state replies → cache update
    end

    App->>Adapter: disconnect()
    Adapter->>Bus: disable_all()
    Adapter->>Bus: close() → stop RX thread

Loading

_{Reviews (4): Last reviewed commit: "converted relateive paths to absolute pa..." | Re-trigger Greptile}