[andrea-cirillo]

Review of the HANDOVER_FW / HANDOVER_APP contract, the app simulator and the firmware, with the fixes.

App (4084a46)

• Simulator in ANGLE/ACRO: the throttle is the collective, as in the firmware (thrust = (stick + 1) / 2, centre = hover). Lift-off happens with the throttle, ANGLE sticks command an angle, ACRO sticks a rate. Before, the throttle was always a climb rate, so the drone never left the ground without TAKEOFF.
• Cockpit safety fix: entering ACRO/ANGLE in the air put the throttle stick at the bottom, which is collective 0 on the real firmware. It now goes to the centre (hover); on the ground it still goes to the bottom.
• HUD: the aircraft symbol was 12 px left of the centre of the pitch ladder, the horizon and the roll pivot. All elements now share one centre, and the roll scale is concentric with the pointer.
• The simulator now sends CHECKS item 10. The ARM throttle rule uses the age of the last RC frame.
• client.test.ts failed in UTC: -0 and 0 differ for Object.is.

Firmware 0.6.1 (b41b801), HANDOVER_FW §6, §7.3, §7.5

1. Stale RC on the ground in ACRO/ANGLE gives an idle collective. Before, a stick left up followed by slide-to-arm spun the motors up.

2. HOLD stick-cancel measures the throttle from its position when HOLD began, not from the centre.

3. After any touchdown the motors stay at idle until the disarm, or until the stick has been at the bottom. Before, an auto LAND with the stick at the centre lifted off again.

4. No altitude estimate:

   • in air is detected from the collective;
   • HOLD holds the hover collective, open loop;
   • LAND descends open loop at 0.85 × hover for at most 20 s.

   Before, the 10 s no-take-off auto-disarm could stop the motors in flight, and LAND kept the last collective.

5. The geofence also applies in ACRO/ANGLE: the climb rate is limited to √(2·2.5·d) below the ceiling.

6. Stick centre = learned hover, so mode changes have no thrust jump.

• Bugs fixed:
  • flash changes are committed only once written, so a failed save leaves RAM unchanged;
  • CAL_SAVE, CAL_RESTORE and SET_NAME no longer write unsaved parameter edits;
  • the ACK cache is reset per link: a BLE re-subscribe no longer makes USB retries execute again.
• Design:
  • CHECKS are built once per 100 ms;
  • STREAM_RATES last one session;
  • prev_auto is reset in fc_init;
  • MOTORS_N is the single source for the motor count (4/6/8, checked at compile time).
• The simulator follows the same rules. HANDOVER_FW.md is updated, including its outdated USB and wall-clock notes, along with flight-core.md.

PC tools (435c3bf)

• Python and MATLAB: since fw 0.6.0 the USB port also streams telemetry. Both download clients extended their wait on any frame, so a lost ACK made them hang. Now only log frames extend the wait. A regression test hangs on the old code.
• MATLAB: acwingLogDownload is at parity with Python:
  • the last flight over BLE;
  • erase, with a typed confirmation;
  • setDTR(true) in the download, the viewer and the capture tools.
• 00 - Matlab/tests/testFlightLog.m runs in MATLAB or Octave. The decoder gives the same numbers as Python on the firmware test vector.

CI (b41b801, 435c3bf, b74bfde)

• New .github/workflows/ci.yml with four jobs:
  • firmware host tests (ASan/UBSan);
  • app (typecheck, tests, build);
  • Python pytest;
  • MATLAB tests in Octave.
• add-to-project.yml retries transient Projects API errors: two runs failed when seven issues were opened together.

Tests

• Firmware host tests: 366 + ext 12 + power 98 + log 63 + link 38 + iekf 32, all passing.
• App: 42 tests, typecheck and build clean.
• Python: 34. Octave: 17.
• The CAL_SAVE rollback, unsaved-parameter and download-timeout tests fail on the old code.

Not verified here: manual checks

• No ARM compiler was available: build for the target, Build firmware 0.6.1 for the target (STM32CubeIDE) #26.
• First run of the CI workflow: First run of the CI workflow (.github/workflows/ci.yml) #27, done by this PR's checks.
• Bench test of the new rules, props off: Bench test (props off) of the 0.6.1 flight-core rules #28.
• Flight validation of the constants, after phase 3: Flight validation of the manual-collective constants (needs phase 3) #29.
• Decision on how the idle after a touchdown is released: Decide: how the idle after a touchdown is released #30.
• MATLAB-only APIs (serialport, setDTR, ble): MATLAB tools with real MATLAB: log download over USB and BLE, DTR #31.
• App on real devices: App on devices: HUD centring, throttle on mode change, simulator in ANGLE/ACRO #32.

---

Generated by Claude Code