Krishna/feat/openarm mini v2 - #3396
Draft
KrishnaH96 wants to merge 48 commits into
Draft
Conversation
Select MIT mode and drain startup traffic before enabling Damiao motors. Send zero-torque OpenYAM frames for compliant encoder readback and expose Viser on all interfaces.
# Conflicts: # CONTEXT.md # dimos/cli/can.py # dimos/cli/dimos.py # dimos/cli/test_can.py # dimos/robot/manipulators/openyam/config.py # pyproject.toml
# Conflicts: # dimos/cli/test_dimos.py
…ody adapter Add OpenArmDamiaoAdapter: both v10 arms (2x DM8006, 2x DM4340, 3x DM4310, send ids 0x01..0x07) and both grippers (DM4310 at 0x08) as one whole-body device over two CAN buses (left=can1, right=can0), with gravity compensation from the bimanual URDF (14 joints, validated order left1..7 then right1..7). Rewrite the OpenArm hardware config on the OpenYAM pattern: one 16-joint WHOLE_BODY component, mock/real selection via global_config.simulation, hardware-measured MIT gains carried over from the legacy adapter, and per-side planning models gaining an explicit coordinator->URDF joint name mapping. Blueprints: coordinator-openarm, openarm-planner-coordinator, keyboard-teleop-openarm and keyboard-teleop-openarm-planner. The keyboard jogs the left arm while the right arm's twist task holds pose; a single servo task drives both grippers, enabled by a new KeyboardTeleopConfig.gripper_joint_names field. The e2e planning-groups test moves to openarm-planner-coordinator since the harness's --simulation flag now selects the in-memory adapter.
The hand-rolled Damiao CAN driver and manipulator-protocol adapter are fully replaced by OpenArmDamiaoAdapter on the whole-body path, which also wires the previously unimplemented grippers. Drop the legacy CAN bring-up scripts (superseded by 'dimos can setup') and the openarm manipulator registry entry.
DM8009 shoulders (DM8006 was a legacy typo), MotorSpecs as plain lists, keyboard teleop module restored to upstream with gripper bindings deferred to a follow-up PR, and dual-arm planning switched to a single bimanual robot model with left_manipulator and right_manipulator groups fed by a hand-written SRDF, since generated SRDFs cannot express cross-robot collision exclusions. Planner blueprint verified against the in-memory adapter in simulation.
Replace the stale v1.0 data package with URDFs generated from the official
OpenArm v2.0 preset pipeline (enactic/openarm_description @ 6c7b720f1ba,
default_bimanual plus per-side pinch gripper presets). Pinch gripper finger
joints are fixed in the generated models so each arm exposes exactly its
seven driven joints while keeping gripper geometry and mass; mesh URIs stay
package relative and ros2_control blocks are dropped. The package ships
v2.0 meshes, the three URDFs, and a PROVENANCE file, and shrinks from 70 MB
to 8 MB.
The v2.0 generator collapses link7, so planning tips move to
openarm_{side}_ee_base_link and the SRDF now disables the sibling finger
pair per hand. Joint naming is unchanged from v1.0, so the adapter topology
and gravity joint order carry over. Verified with pinocchio (14 and 7 DOF
models, finite gravity) and a planner blueprint run in simulation.
RoboPlan generated composite planning groups only for selections spanning two or more robots, so a bimanual robot modeled as one URDF with two planning groups could not plan both arms in a single request. Drop the robot-count restriction; the joint-disjointness requirement and the composite group cap still apply, and overlapping selections are already rejected at the selection layer. Verified in process on the OpenArm 2.0 bimanual model: left, right, and combined fourteen-joint plans all succeed.
Converted OBJ files were named by stem only, so a robot whose visual and collision meshes share a stem (OpenArm v2.0 uses visual/link3.dae and collision/link3.stl) had the collision conversion overwrite the visual one, and viewers rendered collision geometry in visual mode. Suffix the converted name with a hash of the source path and pin the behavior with tests.
Regenerate the v2.0 URDFs with emit_grasp_frame enabled and move the
planning group tip links from the ee flange to openarm_{side}_grasp_frame,
matching the OpenYAM gripper_tip convention. Model DOF, joint order, and
gravity behavior are unchanged; planning verified for left, right, and
combined requests.
The optimistic target ghost rebuilt its per-robot joint values from the current state for every selected group, so with two planning groups on one robot the group processed last discarded the other group's target and the ghost only ever showed one arm's goal. Seed the merge once per robot and overlay each group's target into the same values.
The pinch gripper finger joints are fixed in the generated models and their collision meshes do not intersect at the fixed pose, so the manual exclusions were dead weight. Verified left, right, and combined plans still succeed without the file.
Gripper opening calibrates during activation, and reading it earlier raises, so a connected but not activated whole-body adapter failed the entire state read and read-only bring-up sessions (connect without enable) streamed nothing. Report placeholder gripper states until the adapter is active; arms stream immediately and real openings appear after activation. Found on OpenArm hardware during the read-only phase.
Damiao feedback only updates when the bus is ticked, which the write path does once per control cycle while the adapter is active. A connected but not activated adapter never ticked, so read-only sessions streamed the connect-time snapshot forever. Refresh from the read path whenever the adapter is inactive; the active path is unchanged and still ticks exactly once per cycle.
Ports dimos/teleop/openarm_mini from cc/feat/openarm-mini at 1563ff1, including the multi-turn encoder wrap and the CLI tools. Adapted for the whole-body OpenArm follower: default follower joint names come from openarm_arm_joints, and the sender-side clamp limits mirror the OpenArm v2.0 URDF instead of the v1.0 measured limits.
…llower mini-teleop-openarm drives both arms from two leader ports; the left and right variants drive one side while the other arm holds. Leader joint N maps to follower joint N with no wrist reordering, matching the OpenArm 2.0 conventions on both ends.
Soft arm gains and grippers removed from the topology, the exact configuration validated on both real arms during the leader teleop ladder. Drop or promote this commit before opening the PR.
❌ 1 Tests Failed:
View the top 1 failed test(s) by shortest run time
To view more test analytics, go to the Test Analytics Dashboard |
TomCC7
force-pushed
the
krishna/feat/openarm-damiao
branch
from
August 11, 2026 03:00
fc31ef3 to
fda5f14
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Contribution path
Problem
Solution
How to Test
AI assistance
Checklist