newton-ros-parity

name: newton-ros-parity description: "Validate or replay Moleworks Newton observations through moleworks_ros. Use when bringing up the ROS parity stack, checking /clock and TF, debugging controller-facing terrain topics, replaying Dig3D-style comparisons, or cleaning up stale ROS/Newton processes after parity work."

Newton ROS Parity

Use this skill for Newton <-> moleworks_ros parity and replay workflows. Pair it with ros2-debugging for TF/topic primitives.

Source Of Truth

Start from the active branch docs and scripts:

• docs/ros/WORKTREE_SETUP.md
• docs/ros/DOCKER_RUNBOOK.md
• docs/ROS_PARITY_LOG.md
• scripts/ros/AGENTS.md
• scripts/ros/tests/run_compare_smoke_checks.sh
• scripts/debug/compare_bucket_kinematics_to_pinocchio.py

Open narrower diagnostics only when needed:

• scripts/ros/diagnostics/check_base_pose_match.py
• scripts/ros/diagnostics/dump_synced_ros_snapshot.py
• scripts/ros/diagnostics/dump_newton_height_grid.py
• scripts/ros/diagnostics/dump_gridmap_sampling.py
• scripts/ros/diagnostics/dump_controller_grid_bounds.py

Bringup Preference

If parity uses the running moleworks_ros container:

1. Attach interactively to the container.
2. Create and manage a fresh tmux session inside that shell.
3. Launch components in order instead of blasting everything at once.
4. Inspect pane output between steps with tmux capture-pane.

This avoids fragile nested host quoting and makes cleanup tractable.

Health Checks Before Dig3D / Compare

Use long TF timeouts. Minimum checks:

source /opt/ros/jazzy/setup.bash
source /workspace/moleworks/ros2_ws/install/setup.bash
timeout 15 bash -lc "ros2 topic echo /clock --once"
timeout 15 bash -lc "ros2 topic echo /excavation_mapping/grid_map --once"
timeout 15 bash -lc "ros2 run tf2_ros tf2_echo map CABIN 2>&1" | head -40

Do not start the compare path until /clock, the controller-facing terrain topic, and the required TF chain are live.

Terrain Publisher Rule

• If direct analytical terrain publishing is enabled, keep the standard perception path off unless the user explicitly wants both.
• Do not leave multiple publishers on the controller-facing terrain topic by accident.

Cleanup Is Mandatory

After parity, Dig3D replay, or TorchScript comparison:

1. Kill the tmux session that launched the stack.
2. Scan for orphaned ros2 launch, controller, state-publisher, and helper processes.
3. Kill the exact PIDs that remain.
4. Re-check process lists and host/container resource usage.

Useful checks:

ps -eo pid,cmd | rg 'ros2 launch|dig_3d_controller|robot_state_publisher|mole_joint_state_publisher|publish_flat_excavation_map|compare'
docker stats --no-stream --format '{{.Name}}\t{{.MemUsage}}\t{{.CPUPerc}}' <container>
nvidia-smi --query-gpu=memory.used,memory.total --format=csv,noheader,nounits

Do not leave a stale parity stack running while starting a new one.

Kinematics Contract Notes

When debugging bucket or EE velocity parity:

• State.body_q is the body-frame or body-origin pose.
• State.body_qd is a COM-referenced twist in world coordinates.
• Public eval_jacobian() behaves like a world-origin screw Jacobian in world coordinates.
• Therefore eval_jacobian(...) @ joint_qd is not generally equal to State.body_qd.
• If parity depends on tool-point or COM velocity, transport the screw twist to the correct point or compare against an independent implementation plus finite-difference checks.

Do not interpret a Jacobian mismatch as proof of FK failure without checking the contract first.

Context Hygiene

• Prefer repo scripts and diagnostics over ad-hoc ROS archaeology.
• Keep live CLI output bounded with timeouts and tmux capture-pane.
• If the task turns into prolonged pane/log/process inspection, also use moleworks-subagent-orchestrator.