VTOL Backtransition Improvements

[haumarco]

Solved Problem

The backtransition deceleration controller had two issues:

• Commanded deceleration in the heading direction instead of opposite to velocity direction, causing irrelevant pitch commands when turning during backtransition
• Used unfiltered acceleration measurements, could cause noisy pitch commands

The roll attitude setpoint was binary (100% FW) while the torque blending gradually transitioned from FW to MC based on mc_weight.

Solution

• Correct Deceleration Direction: Changed acceleration direction from heading-based to velocity-based
• Smooth Roll Blending: Roll attitude setpoint now blends based on mc_weight during backtransition
• Acceleration Filtering: Added 50ms alpha filter to reduce noise in pitch commands
• Position-based deceleration only active when position EPH < 10m

Changelog Entry

For release notes:
Bugfix: VTOL backtransition deceleration now uses velocity direction instead of heading
Feature: VTOL backtransition roll attitude blending

• Deceleration now correctly pitches against velocity vector (handles drift/crosswind)
• Roll attitude setpoint smoothly blends between FW and MC controllers
• Added acceleration filtering for improved pitch stability
• Position-based deceleration requires GPS accuracy < 10m

Test coverage

• So far only SITL, will remove draft after flighttests

[github-actions]

🔎 FLASH Analysis

px4_fmu-v5x [Total VM Diff: 256 byte (0.01 %)]

    FILE SIZE        VM SIZE    
--------------  -------------- 
+0.0%    +256  +0.0%    +256    .text
  [NEW]    +456  [NEW]    +456    FlightTaskTransition::computeBackTransitionTiltSetpoint()
   +29%     +68   +29%     +68    FlightTaskTransition::update()
  +7.7%     +64  +7.7%     +64    Standard::update_transition_state()
   +44%     +28   +44%     +28    FlightTaskTransition::activate()
  +7.4%     +16  +7.4%     +16    FlightTaskTransition::FlightTaskTransition()
  +1.1%     +12  +1.1%     +12    matrix::Matrix<>::operator=()
  +3.0%      +8  +3.0%      +8    FlightTask::_evaluateVehicleLocalPosition()
  +0.0%      +8  +0.0%      +8    [section .text]
  [DEL]    -404  [DEL]    -404    FlightTaskTransition::computeBackTranstionPitchSetpoint()
+0.0%    +106  [ = ]       0    .debug_abbrev
+0.0%     +32  [ = ]       0    .debug_aranges
+0.0%    +120  [ = ]       0    .debug_frame
+0.0% +4.75Ki  [ = ]       0    .debug_info
+0.0% +1.19Ki  [ = ]       0    .debug_line
  +300%      +3  [ = ]       0    [Unmapped]
  +0.0% +1.19Ki  [ = ]       0    [section .debug_line]
+0.1% +2.51Ki  [ = ]       0    .debug_loclists
+0.1%    +397  [ = ]       0    .debug_rnglists
+0.0%    +177  [ = ]       0    .debug_str
+0.0%     +48  [ = ]       0    .symtab
  [NEW]     +48  [ = ]       0    FlightTaskTransition::computeBackTransitionTiltSetpoint()
  [DEL]     -48  [ = ]       0    FlightTaskTransition::computeBackTranstionPitchSetpoint()
   +33%     +16  [ = ]       0    Standard::update_transition_state()
   +67%     +32  [ = ]       0    ___Z12get_orb_meta6ORB_ID_veneer
 -40.0%     -32  [ = ]       0    __stm32_configwaitints_veneer
  +1.9%     +32  [ = ]       0    matrix::Matrix<>::operator=()
-2.9%    -256  [ = ]       0    [Unmapped]
+0.0% +9.31Ki  +0.0%    +256    TOTAL

px4_fmu-v6x [Total VM Diff: 256 byte (0.01 %)]

    FILE SIZE        VM SIZE    
--------------  -------------- 
+0.0%    +256  +0.0%    +256    .text
  [NEW]    +456  [NEW]    +456    FlightTaskTransition::computeBackTransitionTiltSetpoint()
   +29%     +68   +29%     +68    FlightTaskTransition::update()
  +7.7%     +64  +7.7%     +64    Standard::update_transition_state()
   +44%     +28   +44%     +28    FlightTaskTransition::activate()
  +7.4%     +16  +7.4%     +16    FlightTaskTransition::FlightTaskTransition()
  +1.1%     +12  +1.1%     +12    matrix::Matrix<>::operator=()
  +3.0%      +8  +3.0%      +8    FlightTask::_evaluateVehicleLocalPosition()
  +0.0%      +8  +0.0%      +8    [section .text]
  [DEL]    -404  [DEL]    -404    FlightTaskTransition::computeBackTranstionPitchSetpoint()
+0.0%    +106  [ = ]       0    .debug_abbrev
+0.0%     +32  [ = ]       0    .debug_aranges
+0.0%    +120  [ = ]       0    .debug_frame
+0.0% +4.75Ki  [ = ]       0    .debug_info
+0.0% +1.19Ki  [ = ]       0    .debug_line
  +300%      +3  [ = ]       0    [Unmapped]
  +0.0% +1.19Ki  [ = ]       0    [section .debug_line]
+0.1% +2.51Ki  [ = ]       0    .debug_loclists
+0.1%    +395  [ = ]       0    .debug_rnglists
  [DEL]      -2  [ = ]       0    [Unmapped]
  +0.1%    +397  [ = ]       0    [section .debug_rnglists]
+0.0%    +171  [ = ]       0    .debug_str
+0.0%     +48  [ = ]       0    .symtab
  [NEW]     +48  [ = ]       0    FlightTaskTransition::computeBackTransitionTiltSetpoint()
  [DEL]     -48  [ = ]       0    FlightTaskTransition::computeBackTranstionPitchSetpoint()
   +33%     +16  [ = ]       0    Standard::update_transition_state()
  +1.9%     +32  [ = ]       0    matrix::Matrix<>::operator=()
-4.9%    -256  [ = ]       0    [Unmapped]
+0.0% +9.30Ki  +0.0%    +256    TOTAL

Updated: 2025-10-03T14:13:08

[mbjd]

I like it generally.

• I fully support the change from pitch to tilt setpoint, I see it as a clear improvement in the edge case of (very) bad yaw tracking in backtransitions.
• The roll setpoint which is now continuously blended from FW to MC looks to me like an improvement from a gut feeling perspective, but it's hard to predict if that's actually going to translate to better real world behaviour. Am excited for the test results, especially with crosswind :)
• I presume you have cases where without the additional alpha filter the pitch / elevator setpoints would be unacceptably noisy -- If not strictly necessary, I would vote against the alpha filter to keep complexity from increasing.