RobotGuide

Engineering practice · topic F6 · 30 hours

System integration & field commissioning

Between «runs in the lab» and «works in the field» lie weeks. Integration skill is the most underrated hard skill.

Engineering stage · Hardcore
Algorithm · Aware System software · Proficient Embedded & controls · Hands-on Mechanical · Hands-on

Tools
Tools

What to learn

  • Integration order: single module → pairwise interfaces → subsystems → whole machine
  • Controlling field variables: environment, power supply, network, floor material
  • Delivery checklist and acceptance criteria (who signs, judged by what)
  • Rollback plan: how to recover within ten minutes of a bad change

The lesson · plain words

System integration is the last mile from «every module works on its own» to «the whole machine runs», and the most grinding stretch: modules that are all green in unit tests will fight once assembled. This page covers the discipline of bring-up.

Five bring-up levels: never enter the next until this one is green

Look at the figure's five levels: power → comms → drives → sensors → closed loop. Each level has explicit acceptance criteria (voltage ripple, heartbeat frequency, encoder counts, frame rate, unloaded jogging). Skipping levels = two unknowns fighting — you do not know which one to fix.

Leave one «one-click self-check» command per level (a script printing PASS/FAIL) and rerun them in order after any field reboot. This one discipline cuts field troubleshooting from half a day to half an hour.

Interactive figure · drag it and see

What this figure shows: a bring-up checklist with five levels — ① power (rail voltages, ripple), ② comms (CAN / Ethernet heartbeat), ③ drives (motor enable, encoder counts), ④ sensors (camera frames, IMU rate), ⑤ closed loop (dry-run jog, limit protection) — each with its acceptance criterion, a current-stage indicator, and a one-line self-check per level. Takeaway: bring-up discipline is never entering the next level until this one is green; skipping levels means two unknowns fighting, and safety nets (e-stop, limits, software clamps) land before everything else.

Safety nets land first

E-stop, limit switches, and software clamps must be in place before «closed-loop bring-up» — they are the safety rope for all debugging that follows. The first power-up must support one-key disable; the first motion must happen at 10% of the clamped speed.

Field troubleshooting priority: safety first, then power, then comms, and only then algorithms. Reversing the order usually costs you a board or an injured hand.

Documentation discipline during integration

Three tables travel with the code: wiring table (which device into which port), parameter table (each node's key parameters and defaults), version table (firmware/code version of every module). Swap a sensor model in the field without a wiring table and you are doing archaeology.

Log discipline: collect whole-machine logs centrally (D3's journalctl + ROS log), timestamped and aligned. Post-mortem analysis depends on «what every module was doing at the same moment»; scattered logs are as good as none.

How deep to go

You can write a bring-up checklist (five levels + criteria + self-check commands); build one-key disable and software clamps; and complete a first bring-up on a new machine by the discipline. The market premium for systems integration engineers comes precisely from this replicable discipline.

Three questions — do you really get it?

1. What is the main risk of skipping bring-up levels (say, tuning the closed loop directly)?

  • ASlower progress
  • BSeveral unknowns coupled together — you cannot localize which level is at fault
  • CNo risk
Show answer

B. Each level is verified independently before the next, so faults stay unit-localized.

2. At what stage should limit protection and software clamps be in place?

  • ARetrofitted after all tuning is done
  • BBefore closed-loop bring-up
  • CBefore delivery
Show answer

B. They are the safety rope of the debugging stage. They must be active before the first motion.

3. What is the correct priority order for field troubleshooting?

  • AAlgorithm → comms → power
  • BSafety → power → comms → algorithm
  • CWhatever comes to mind
Show answer

B. Check bottom-up. The algorithm is the last thing to suspect (F5's principle: suspect the most probable first).

Keep going deeper

Links without a marker stay on this site; the rest point to the Chinese edition until its translation batch lands.

How it fails in the field

  • Starting integration at the whole-machine level; when something breaks, nobody knows whose fault it is and you eliminate modules one by one
  • Field parameter changes with no record; back at the office nothing can be reproduced

You have learned enough when

You can write an integration plan (order, acceptance criteria, and rollback included) and complete a whole-machine bring-up by it.

The Chinese edition carries the full topic page for System integration & field commissioning, including the complete outline, depth requirements per career track, and further derivations: robotguide.dev/k/F6/.