Pr 4631 ported 4575 rc9

[vigneshm-debug]

https://app.notion.com/p/airbound/Airbound-4-6-3-1-RC-8-39e21adf4be980159aefd67d222c4ece?source=copy_link

[greptile-apps]

Greptile Summary

This PR ports rc8/rc9 Airbound features onto the ArduPlane 4.6.3 base, combining the disarm watchdog (DARM_WDG_T), a parameterized land-detector timeout (LND_DET_TIM), a full rewrite of tailsitter attitude relaxation with hysteresis, new Lua scripting bindings for attitude telemetry, and updates to the LTE modem and autobailout scripts.

• Attitude relaxation rewrite (AC_AttitudeControl): replaces the old single-threshold approach with a static-local LPF and hysteresis pair (RELX_HI/RELX_LO); also adds a hard pitch-clip (PIT_CLIP_MAX). Previously flagged issues (static locals surviving disarm, member/local variable overwrite in thrust-error branches) remain unresolved.
• Landing detection rework (quadplane.cpp): introduces DARM_WDG_T watchdog and LND_DET_TIM; the 200 ms default is far below the declared @Range floor and reduces both check_land_complete and check_land_final windows dramatically from their prior hard-coded 4000/6000 ms values.
• Scripting and Lua (lua_bindings.cpp, autobailout.lua): adds four new C-side bindings returning Lua tables, but bindings.desc/docs.lua still declare three separate return values per call.

Confidence Score: 2/5

Not safe to merge; multiple control-path defects in the attitude controller and landing detection remain unaddressed, and the newly added weathervane gain reduction in QPOS_POSITION2 silently has no effect.

The attitude controller thrust-error branches write to _ang_vel_body (member) then unconditionally overwrite it with the untouched local, discarding yaw-gyro clamps and partial feedforward every high-tilt cycle. Land-detector timeouts collapse to 200 ms, opening a false-landing window during powered descent. The new weathervane->set_gain(wvane_max_gain/3) in vtol_position_controller is overwritten every frame by tailsitter.output(), so the intended gain reduction never applies.

AC_AttitudeControl.cpp (thrust-error branch variable scoping and static local state), quadplane.cpp (land detector timeout and weathervane gain calls at lines 2880/2891)

Important Files Changed

Filename	Overview
ArduPlane/quadplane.cpp	Adds DARM_WDG_T watchdog, LND_DET_TIM parameter, and weathervane gain changes for tailsitter QPOS_POSITION2; the set_gain call is immediately overridden by tailsitter output, and the 200 ms land-detector timeout carries multiple previously-flagged regressions.
libraries/AC_AttitudeControl/AC_AttitudeControl.cpp	Rewrites attitude relaxation with hysteresis and pitch-clip; introduces static local relaxation state and a member-vs-local overwrite in thrust-error branches that discards yaw and feedforward corrections (previously flagged P1).
libraries/AC_AttitudeControl/AC_AttitudeControl.h	Replaces single-threshold relaxation params with hysteresis pair and adds _att_max_pit; removes ts_back_transition_done and LAND*_MULT fields.
ArduPlane/tailsitter.cpp	Removes HAL_LOGGING_ENABLED guard from WriteStreaming call (previously flagged); otherwise no functional changes.
libraries/AP_HAL_ChibiOS/hwdef/Pixhawk6C-bdshot/scripts/LTE_modem.lua	Refactors signal parsing into QENG/QCSQ handlers; QCSQ sends RSSI/RSRP/RSRQ unconditionally while CPSI still gates on OPT.SIGNALS.
libraries/AP_HAL_ChibiOS/hwdef/Pixhawk6C-bdshot/scripts/autobailout.lua	Adds rolling-window pitch averaging and rate-based prediction; is_vtol_flight implicit global and multiple para_deploy implicit globals remain outstanding.
libraries/AP_Scripting/lua_bindings.cpp	Adds four new C-side bindings returning Lua tables; bindings.desc still declares 3 returns per call (mismatch previously flagged).

Sequence Diagram

sequenceDiagram
    participant PC as vtol_position_controller()
    participant AC as attitude_controller_run_quat()
    participant TS as tailsitter.output()
    participant WV as AC_WeatherVane

    PC->>WV: set_gain(wvane_max_gain/3) [QPOS_POSITION2]
    Note over PC,WV: Intended: reduce weathervane to 1/3

    AC->>AC: compute ang_vel_body (local)
    alt "thrust_error > 3x threshold"
        AC->>AC: "_ang_vel_body.z = gyro.z (member)"
        AC->>AC: reset_I()
    else "thrust_error > 2x threshold"
        AC->>AC: "_ang_vel_body += feedforward (member)"
    else normal
        AC->>AC: "ang_vel_body += feedforward (local)"
    end
    AC->>AC: "_ang_vel_body = ang_vel_body (local overwrites member!)"

    TS->>WV: set_gain(wvane_max_gain) [full pitch-based gain]
    Note over TS,WV: Overrides vtol_position_controller set_gain

    WV->>WV: get_yaw_out() uses full gain, not 1/3

Loading

_{Reviews (5): Last reviewed commit: "Remove debug LTE_QENG named-float broadc..." | Re-trigger Greptile}

[greptile-apps]

static locals carry stale relaxation state across flights

relaxation_factor_lpf and _att_relax_active are static local variables, so they survive arm/disarm cycles without any reset. If the vehicle was in a high-tilt state when power is cycled or an emergency disarm occurs, relaxation_factor_lpf can be near 1.0 on the next arm. The attitude controller would then immediately zero out the pitch setpoint (euler_sp.y *= (1.0f - 1.0f)) at the start of the next flight until the LPF decays, which could cause unexpected control behaviour during takeoff. These should be promoted to member variables in AC_AttitudeControl so they can be properly reset (e.g., in a reset or initialise method).

[greptile-apps]

Debug bypass of OPT.SIGNALS gate committed to main branch

The OPT.SIGNALS guard that previously controlled whether extended signal metrics (LTE_RSRP, LTE_RSRQ, LTE_SINR, LTE_BAND, LTE_CID, LTE_MCCMNC) are broadcast was removed. The in-code comment explicitly states: "Revert this gating once you've confirmed values are reaching MP." Broadcasting these named floats unconditionally on every +QENG parse adds unnecessary GCS telemetry traffic in production and cannot be disabled via LTE_OPTIONS. The original gate should be restored.

[greptile-apps]

is_vtol_flight is assigned without local, making it an implicit global variable in Lua. It should be declared locally to avoid polluting the global namespace and to prevent potential cross-script interference.

Suggested change

	is_vtol_flight = in_vtol_flight()
	local is_vtol_flight = in_vtol_flight()

[greptile-apps]

LND_DET_TIM default (200 ms) is outside its own documented @Range (1000–4000)

The parameter annotation states @Range: 1000 4000 but the default and the defaults.parm entry are both set to 200 ms. The @Range is advisory in ArduPilot (not enforced), so this won't crash, but it misleads ground station users who inspect the parameter limits and may cause confusion when tuning. Either update @Range: 1 4000 to reflect the actual use, or raise the default to the minimum documented value.

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!

[greptile-apps]

The #if HAL_LOGGING_ENABLED guard was removed from this WriteStreaming call. On targets where HAL_LOGGING_ENABLED is not defined the call is either a no-op stub or fails to compile, depending on the platform. The original guard was deliberate; restore it to preserve portability and avoid unconditional overhead.

Suggested change

	float position_pitch_sp = quadplane.pos_control->get_pitch_cd();
	// Add logging for desired thrust vectoring angles
	AP::logger().WriteStreaming("PHID", "TimeUS,DesL,DesR,AhrsPitch,WVGain,WVGainS",
	"sddd--", // seconds, degrees
	"F00000", // micro (1e-6), no mult (1e0)
	"Qfffff", // uint64_t, float
	AP_HAL::micros64(), tilt_left/100, tilt_right/100,pitch_cd/100,weathervane_gain,gain_slope);
	#endif
	AP::logger().WriteStreaming("PHID", "TimeUS,DesL,DesR,AhrsPit,WVGain,VHPwEe,PosPit",
	"sddd--d", // seconds, degrees
	"F000000", // micro (1e-6), no mult (1e0)
	"Qffffff", // uint64_t, float
	AP_HAL::micros64(),tilt_left/100,tilt_right/100,pitch_cd/100,weathervane_gain,extra_elevator/100,position_pitch_sp/100);
	float position_pitch_sp = quadplane.pos_control->get_pitch_cd();
	#if HAL_LOGGING_ENABLED
	// Add logging for desired thrust vectoring angles
	AP::logger().WriteStreaming("PHID", "TimeUS,DesL,DesR,AhrsPit,WVGain,VHPwEe,PosPit",
	"sddd--d", // seconds, degrees
	"F000000", // micro (1e-6), no mult (1e0)
	"Qffffff", // uint64_t, float
	AP_HAL::micros64(),tilt_left/100,tilt_right/100,pitch_cd/100,weathervane_gain,extra_elevator/100,position_pitch_sp/100);
	#endif

[greptile-apps]

Debug CSQ GCS text print committed without a removal plan

The comment explicitly marks this as -- DEBUG: and it throttle-prints LTE CSQ: RSSI=… BER=… to the GCS text channel every 5 s regardless of LTE_OPTIONS. Unlike named floats, GCS text messages are visible in all ground stations and consume MAVLink bandwidth continuously. This should be guarded behind OPT.LOGALL or removed before merging.

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!

[greptile-apps]

QPOS_LAND_COMPLETE is unreachable after a soft disarm

The new early return exits check_land_complete() when !motors->armed(). This means that if the motors are already disarmed (e.g. after a quiet-disarm command, an arming watchdog trip, or the watchdog block just above), the position controller can never transition to QPOS_LAND_COMPLETE. Any logic that gates on QPOS_LAND_COMPLETE (logging, mission continuation, payload-place handling) would then be skipped permanently for that landing.

[greptile-apps]

check_land_final timeout silently dropped from 6000 ms to 200 ms

The in-code comment on the block immediately above (/* also apply landing detector, in case we have landed in descent phase. Use a longer threshold */) explicitly acknowledges that check_land_final must use a longer threshold than check_land_complete. Before this PR, check_land_final called land_detector(6000) and check_land_complete called land_detector(4000). After the refactor both call the parameterless land_detector(), which uses landing_detect.timeout_ms — defaulting to 200 ms. The lower-limit guard consequently shrinks from 7 000 ms to 1 200 ms. During the powered-descent phase (QPOS_LAND_DESCEND), any 200 ms window with less than detect_alt_change altitude variation could trigger a false land-complete, causing premature transition and potential crash.

[greptile-apps]

Docs and bindings descriptor claim 3 return values; implementation returns 1 table

bindings.desc registers qp_att_desired, qp_att_actual, and qp_angle_rate with 0 3 (0 args, 3 returns). docs.lua annotates them the same way — three separate @return number entries. But each C implementation pushes a single Lua table and returns 1. A script author following the documented API would write local roll, pitch, yaw = qp_att_desired() and silently get (table, nil, nil). The Lua scripts in this PR that use these functions do so correctly (capturing the single table), but the discrepancy between the declared and actual return shape will mislead any future scripts relying on the documentation.