feat(go2): unitree-go2-joystick blueprint, operator input recorded and plotted

[KrishnaH96]

Problem

The viewer's WASD input is converted to tele_cmd_vel right away and never published in raw form, so --record cannot store it, and no Go2 blueprint shows the operator's input next to the velocity it turns into.

Solution

• RerunWebSocketServer publishes the viewer's twist message as joystick: Out[Joy] before converting it to tele_cmd_vel. 6 added lines, the existing conversion untouched.
• unitree-go2-joystick: unitree-go2 with its own viewer window adding joystick, tele_cmd_vel and cmd_vel plots under the camera and odom, one color per axis, step interpolation. Run with --record to store all three. The default unitree-go2 window is unchanged.
• unitree-go2-joystick-replay: the same window on the replay module from feat: replay blueprint - dimos --replay-db <memory.db> run replay #3738, for recordings without a run folder such as an LFS dataset.
• LFS dataset go2_teleop_sf_office_2026-09-18 (379 MB): 3 min real Go2 run, WASD driving then two click-to-go goals.

+237 lines, 0 deletions. A recording made with unitree-go2-joystick lands in a run folder named after it, so plain run replay finds this window automatically.

Note: dimos-viewer converts keys to velocities itself, so joystick equals tele_cmd_vel today. tele_cmd_vel vs cmd_vel is the pair that diverges, whenever nav drives (2:22 and 2:43 in the clip).

How to Test

Replay the shared recording, no robot needed. The dataset is pulled from LFS the first time:

dimos --replay-db go2_teleop_sf_office_2026-09-18 run unitree-go2-joystick-replay

Screencast_2026-09-18_14-50-29.mp4

To run the blueprint yourself and collect data on hardware: dimos --record run unitree-go2-joystick --robot-ip <ip>.

Alternative

#4224 does the same thing by adding the plots to the default unitree-go2 window, so every Go2 user sees them, plus a unitree-go2-replay blueprint for LFS datasets. Both are hardware-tested and mergeable; @spomichter pick one and I will close the other.

AI assistance

Claude Code (Claude Opus 5) helped survey the code, draft the change and run the checks. I reviewed and understand every line.

Checklist

• I have read and approved the CLA.

[codecov]

Codecov Report

❌ Patch coverage is 96.29630% with 3 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...itree/go2/blueprints/smart/unitree_go2_joystick.py	90.90%	2 Missing and 1 partial ⚠️

@@           Coverage Diff           @@
##             main    #4242   +/-   ##
=======================================
  Coverage   79.72%   79.72%           
=======================================
  Files        1508     1510    +2     
  Lines      142607   142688   +81     
  Branches    12209    12211    +2     
=======================================
+ Hits       113689   113763   +74     
- Misses      25478    25485    +7     
  Partials     3440     3440           

Components	Coverage Δ
Tests	95.42% <100.00%> (+<0.01%)	⬆️

Flag	Coverage Δ
OS-ubuntu-24.04-arm	75.36% <96.29%> (+<0.01%)	⬆️
OS-ubuntu-latest	76.18% <96.29%> (+0.01%)	⬆️
Py-3.10	75.95% <96.29%> (+0.01%)	⬆️
Py-3.11	76.15% <96.29%> (+<0.01%)	⬆️
Py-3.12	76.16% <96.29%> (+0.01%)	⬆️
SelfHosted-Large	30.63% <50.61%> (+0.01%)	⬆️
SelfHosted-Linux	36.66% <50.61%> (+0.01%)	⬆️
SelfHosted-macOS	36.99% <50.61%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
dimos/robot/all_blueprints.py	100.00% <ø> (ø)
.../go2/blueprints/smart/test_unitree_go2_joystick.py	100.00% <100.00%> (ø)
dimos/visualization/rerun/test_websocket_server.py	98.49% <100.00%> (+0.12%)	⬆️
dimos/visualization/rerun/websocket_server.py	86.36% <100.00%> (+0.53%)	⬆️
...itree/go2/blueprints/smart/unitree_go2_joystick.py	90.90% <90.90%> (ø)

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[greptile-apps]

Confidence Score: 4/5

The verified test-coverage gap is non-blocking, but it should be fixed so Linux SocketCAN behavior remains covered without optional macOS-only dependencies.

What we checked:

• T-Rex produced a proof for a posted P2 finding.
• T-Rex authored a focused harness for the Go2 joystick blueprint entries and attempted to run it with the Unitree dependency extra, but the build stopped due to missing portaudio.h.
• T-Rex documented the blocker caused by missing portaudio.h when pyaudio==0.2.14 is pulled for dimos[unitree], noting that no source code was modified and only the evidence harness was authored.

Summary

• Adds Go2 joystick telemetry publication plus live and replay viewer layouts for comparing operator input with teleoperation and final velocity commands.
• Registers the new Go2 blueprints and adds a replay dataset pointer.
• Linux CAN command coverage is accidentally disabled whenever the optional control dependency is not installed.

T-Rex validation blocked

• The Go2 joystick blueprint could not be imported or rendered because the Unitree dependency requires unavailable PortAudio headers. The replay dataset also could not be fetched because Git LFS is unavailable.

_{Reviews (1) · Last reviewed commit: "data: go2_teleop_sf_office_2026-09-18 re..."}

[greptile-apps]

Comments Outside Diff

These findings sit on lines the diff does not cover, so they could not be posted inline. Each one leaves this list once its file changes.

•  Linux CAN tests skipped dimos/cli/test_can.py:22 ▶

  The module-level optional-dependency skip prevents every CAN CLI test from running when the control extra is absent, including Linux SocketCAN paths that do not import that dependency. This is non-blocking, but it removes regression coverage for Linux device discovery, setup, status, up, and down commands in the default environment.

  Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!