Skip to content

usage.md

Latest commit

 

History

History
156 lines (108 loc) · 3.49 KB

usage.md

File metadata and controls

156 lines (108 loc) · 3.49 KB
Raw

Usage

Last updated: 2026-07-22, Asia/Shanghai.

This document is the operator-level workflow across the three project components. Component-specific details live in each thirdparty repository.

1. Start Or Check Collector

On the Collector PC:

ssh lvjun@10.128.1.95
cd /ssd1/shenyibo/Quest3DataCollector
pc/offline_calibration/scripts/start_lab_receiver.sh --restart

From this workstation, check:

curl.exe -s -o NUL -w "%{http_code}\n" http://10.128.1.95:8765/

Expected result: 200.

Viewer:

  • http://10.128.1.95:8765/
  • http://10.128.1.95:8765/recordings

2. Quest Recording

Quest app package:

com.Apricity.EyeTrackingTest

Check device:

adb devices

Controls:

  • A button: start/stop formal recording telemetry.
  • B button: start/stop PC calibration recording.
  • Right hand side/grip trigger: enable right-controller robot teleop during formal robot recording.
  • Right index trigger: gripper control when gripper support is enabled.

3. Calibration

Quest B-button calibration sends HTTP samples to:

http://10.128.1.95:9101

Collector raw calibration folders:

pc/offline_calibration/raw/<recordId>/

Collector calibration outputs:

pc/offline_calibration/outputs/pc_live_calibration/<recordId>/

4. Formal Recording Artifacts

Formal A-button recordings are written under:

pc/offline_calibration/pc_recordings/<recordId>/

Important files:

  • pc_telemetry_raw.jsonl
  • pc_samples.jsonl
  • pc_controllers.csv
  • pc_session_summary.json
  • robot_realsense/robot_states.jsonl
  • robot_realsense/controller_motion.jsonl
  • robot_realsense/video_frames.jsonl
  • robot_realsense/gripper_commands.jsonl
  • robot_realsense/samples.jsonl, the fixed 30 Hz training-alignment stream
  • teleop_latency_analysis.json, when teleop latency analysis has run

Rate contract:

  • robot_realsense/robot_states.jsonl: 90 Hz Flexiv state.
  • RealSense video: 30 Hz per role.
  • robot_realsense/samples.jsonl: stable 30 Hz fused timeline.

Performance audit:

.venv312/bin/python pc/offline_calibration/scripts/quest_pc_receiver.py audit-performance \
  --pc-session pc/offline_calibration/pc_recordings/<record_id>

5. Replay And Teleop Latency

Open:

http://10.128.1.95:8765/recordings

Select a record to inspect Quest samples, gaze, controller poses, robot TCP, camera videos, gripper events, and teleop latency.

Manual teleop latency analysis:

cd /ssd1/shenyibo/Quest3DataCollector
.venv312/bin/python pc/offline_calibration/scripts/quest_pc_receiver.py teleop-latency \
  --pc-session pc/offline_calibration/pc_recordings/<record_id>

Interpretation:

  • targetToRobot is preferred. It compares Collector command target TCP speed to Flexiv robot TCP speed on the same PC timeline.
  • controllerToRobot is a fallback. It compares Quest right-controller motion to Flexiv robot TCP speed.
  • Positive lag means robot motion follows the command/controller by that delay.
  • This is a system responsiveness metric, not absolute one-way network latency.

6. Training Handoff

Current Gaze-WAM training deployment:

ssh H200-4042
cd /mnt/workspace/shenyibo/gaze-wam
. .venv/bin/activate

Current missing bridge:

Collector pc_recordings/<recordId>/ -> canonical Gaze-WAM robot zarr

Do not lock converter semantics without confirming:

  • policy camera source and serial
  • action_abs_tcp target definition
  • gripper label source
  • timestamp alignment strategy