Skip to content

feat(falcon): v0.19.4 — closed-loop cascade wired into gz-sim; tuning gap surfaced #49

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
avrabe merged 1 commit into from
May 27, 2026
Merged

feat(falcon): v0.19.4 — closed-loop cascade wired into gz-sim; tuning gap surfaced #49

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
94 changes: 93 additions & 1 deletion artifacts/features/FEAT-FALCON-rollout.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1753,6 +1753,98 @@ artifacts:
- type: depends-on
target: FEAT-FALCON-v0.19.2

- id: FEAT-FALCON-v0.19.4
type: feature
title: "v0.19.4 — closed-loop cascade wired into gz-sim; tuning gap surfaced"
status: approved
description: >
LANDED. The verified cascade (relay-ekf → relay-pos → relay-att
→ relay-rate → relay-mix-quad) is now wired into falcon-sitl-gz
and runs end-to-end against real gz Harmonic physics. First
time the safety-critical control surface is in the loop under
real physics. Verdict is FAIL on hover-stability — body stays
on the ground in mixer torque-saturation mode — but the
architectural deliverable is the wiring + the falsifiable
failure mode that v0.19.5 picks up.

Bench result (15-second hover, setpoint NED (0,0,-2)):
verdict: FAIL
final_dist=2.44m peak_dist=3.30m rms_steady=2.16m
min_dist=1.75m
counters: imu_recv=3526 (203 Hz) navsat_recv=870 (50 Hz)
motor_send=1500 (100 Hz)

Cascade pattern matches falcon-sitl-hover::run_mission:
physics.measure → EKF → POS → ATT → RATE → MIX → physics.step.

Ships:
- examples/falcon-sitl-gz/src/main.rs::run_closed_loop_hover
— full cascade tick.
- --scenario=hover (default) → closed-loop;
--scenario=open-loop-climb → v0.19.3's constant 70 % PWM
(preserved as wire-level smoke test).
- SDF tweaks: NavSat 10 → 50 Hz (so finite-diff velocity has
rate); body spawn 0.1 → 1.0 m.
- cargo test `closed_loop_hover_compiles_and_ticks_on_mock`
— panic + NaN-free contract.
- bench-evidence/gz-sim/2026-05-27-v0.19.4-*.md +
1779904889-gazebo-hover-{harness.log,ticks.csv}.
- FV-FALCON-SIM-010 records the wiring + falsifiable failure.

Falsifiable failure mode:
Body sits at d≈0 with motors in bang-bang
[1.00, 0.00, 0.00, 0.00] torque saturation. PosController
commands large attitude tilt → AttController large rate →
RatePid large torque → QuadMixer's priority-preserving
saturation (PX4-style: yaw last, then thrust) starves
thrust. Net: rotors spin asymmetrically, zero net thrust →
body never lifts.

What v0.19.4 closes:
- v0.19.3's deferred "scenario=hover swap to closed-loop".
- "Verified cascade is integrable into real-physics
simulator" — proven by 1500 ticks of stable execution.

v0.19.5 candidates (each separable):
- Re-tune PosController gains for 700 g falcon-quad
(defaults assume ~1 kg).
- Bring back decimation pattern (RATE 1 kHz, ATT 250 Hz,
POS 50 Hz) by bumping harness tick rate.
- Use gz body-velocity topic OR fuse IMU+NavSat in
relay-ekf rather than finite-diff NavSat.
- Add a minimum-hover-thrust floor to QuadMixer (Verus
contract update).

Verification:
- cargo test -p falcon-sitl-gz → 8/8 (was 7/7; +1 closed-
loop smoke).
- cargo build --features gazebo → green.
- cargo run + gz sim → cascade runs 1500 ticks, no NaN,
body responds (min_dist=1.75 m), doesn't settle.
- rivet validate → PASS.

Honestly NOT claimed:
- That the cascade flies stably. Body is on the ground.
"Falcon flies stably" is the v0.19.x finish line, not
v0.19.4.
- That failure mode is Verus-grade diagnosed. Mixer-
saturation hypothesis is consistent with bang-bang motor
pattern; tick-by-tick mixer I/O trace is v0.19.5 work.
tags: [falcon, milestone, v0.19.4, gazebo, closed-loop, bench-evidence, landed]
fields:
release-target: "closed-loop cascade wired; tuning gap surfaced"
bench-date: "2026-05-27"
gz-version: "8.11.0"
counters-observed: "imu_recv=3526 navsat_recv=870 motor_send=1500"
final_dist_m: 2.437
rms_steady_m: 2.156
verdict: FAIL
failure-mode: "mixer priority-preserving saturation starves thrust under aggressive torque"
cascade-components: ["relay-ekf", "relay-pos", "relay-att", "relay-rate", "relay-mix-quad"]
links:
- type: depends-on
target: FEAT-FALCON-v0.19.3

- id: FEAT-FALCON-v1.0
type: feature
title: "v1.0 — six-domain credit dossier + airframe variants"
Expand Down Expand Up @@ -1789,4 +1881,4 @@ artifacts:
- type: implements
target: SYSREQ-FALCON-010
- type: depends-on
target: FEAT-FALCON-v0.19.3
target: FEAT-FALCON-v0.19.4
103 changes: 103 additions & 0 deletions artifacts/verification/FV-FALCON-SIM-010.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,103 @@
artifacts:
- id: FV-FALCON-SIM-010
type: sw-verification
title: "v0.19.4 — closed-loop cascade wired into falcon-sitl-gz; tuning gap surfaced"
status: approved
description: >
v0.19.4 wires the verified cascade (relay-ekf → relay-pos →
relay-att → relay-rate → relay-mix-quad) into the falcon-sitl-gz
bridge and runs it end-to-end against real gz Harmonic physics.
First time the safety-critical control surface is in the loop
under real physics. Verdict is FAIL on hover-stability — body
stays on the ground in mixer torque-saturation mode — but the
architectural deliverable is the wiring + the falsifiable failure
mode that v0.19.5 picks up.

Bench result (15-second closed-loop hover, setpoint NED (0,0,-2)):
verdict: FAIL
final_dist=2.44m peak_dist=3.30m rms_steady=2.16m min_dist=1.75m
counters: imu_recv=3526 (203 Hz) navsat_recv=870 (50 Hz) motor_send=1500 (100 Hz)

Cascade pattern (mirrors falcon-sitl-hover::run_mission):
1. physics.measure() → (ImuSample, position_ned)
2. Ekf::tick(imu_sample) → attitude estimate
3. PosController::tick(pos_ned, v_ned_finite_diff, est_quat,
setpoint) → attitude-setpoint + thrust
4. AttController::tick(est_quat, att_setpoint) → rate-setpoint
5. RatePid::tick(gyro, rate_setpoint) → torque
6. QuadMixer::mix(torque, thrust) → 4× motor PWM
7. physics.step(motors, dt) → bridge → gz

v0.19.4 ships:
- examples/falcon-sitl-gz/src/main.rs::run_closed_loop_hover —
full cascade tick.
- --scenario=hover (default) → closed-loop.
- --scenario=open-loop-climb → v0.19.3's constant 70 % PWM
(preserved as a wire-level smoke test).
- SDF tweaks: NavSat 10 → 50 Hz; body spawn 0.1 → 1.0 m.
- New test `closed_loop_hover_compiles_and_ticks_on_mock`
pins panic-free + NaN-free contract on the mock backend.
- bench-evidence/gz-sim/2026-05-27-v0.19.4-*.md +
1779904889-gazebo-hover-{harness.log,ticks.csv}.

Failure-mode diagnosis (falsifiable):
- Body sits on the ground (d ≈ 0) with motors in bang-bang
[1.00, 0.00, 0.00, 0.00]-style torque saturation.
- PosController commands large attitude tilt → AttController
large rate → RatePid large torque → QuadMixer's
priority-preserving saturation sacrifices THRUST to torque
(per crates/relay-mix-quad's PX4-style behaviour: yaw last,
then thrust, when motors clip).
- Net: rotors spin asymmetrically with zero net thrust → body
stays on the ground.
- This is a TUNING gap, not a wiring gap. falcon-sitl-hover's
pure-Rust toy-plant converges in 4 s; gz delta is physics +
finite-diff velocity + controller-tick-rate mismatch.

v0.19.5 candidates (each separable):
- Re-tune PosController gains for 700 g falcon-quad (defaults
assume ~1 kg).
- Bring back the decimation pattern (RATE 1 kHz, ATT 250 Hz,
POS 50 Hz) by bumping harness tick rate.
- Use gz's body-velocity topic OR Kalman-fuse IMU+NavSat in
relay-ekf rather than NavSat finite-diff.
- Add a "minimum hover thrust" floor to QuadMixer (Verus
contract update).

Tests:
- cargo test -p falcon-sitl-gz → 8/8 (was 7/7 at v0.19.3;
+1 closed_loop_hover_compiles_and_ticks_on_mock).
- cargo build --features gazebo → green.
- cargo run + gz sim bench → cascade runs to completion, no
NaN, body responds (min_dist=1.75 m) but doesn't settle.

Honestly NOT claimed:
- That the cascade flies stably. v0.19.3 was open-loop
ballistic ascent; v0.19.4 wires safety but doesn't hover.
- That the failure mode is fully Verus-grade diagnosed.
The mixer-saturation hypothesis is consistent with the
bang-bang motor pattern; tick-by-tick mixer I/O trace is
v0.19.5 verification work.

What v0.19.4 actually closes:
- The v0.19.3 deferred "scenario=hover swap to closed-loop".
- "The verified cascade is integrable into a real-physics
simulator" — proven by 1500 ticks of stable execution.
tags: [falcon, sim, gazebo, bench-evidence, closed-loop, tuning-gap, v0.19.4]
fields:
bench-evidence-dir: bench-evidence/gz-sim/
bench-date: "2026-05-27"
gz-version: "8.11.0"
counters-observed:
imu_recv: 3526
navsat_recv: 870
motor_send: 1500
final_dist_m: 2.437
rms_steady_m: 2.156
min_dist_m: 1.750
verdict: FAIL
failure-mode: "mixer priority-preserving saturation starves thrust under aggressive torque demand"
cascade-components: ["relay-ekf", "relay-pos", "relay-att", "relay-rate", "relay-mix-quad"]
links:
- type: verifies
target: SWREQ-FALCON-SIM-P04
15 changes: 15 additions & 0 deletions bench-evidence/gz-sim/1779904889-gazebo-hover-harness.log
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
falcon-sitl-gz bench-evidence
backend: gazebo
scenario: hover
timestamp: 1779904889

steps: 1500
final_dist: 2.437 m
peak_dist: 3.300 m
rms_steady: 2.156 m (last 5 s)
min_dist: 1.750 m
wall: 17.502 s
imu_recv: 3526
navsat_recv: 870
motor_send: 1500
verdict: FAIL
Loading
Loading