feat(falcon): v0.19.3 — first PASS verdict under real gz physics (cascade flies)

artifacts/features/FEAT-FALCON-rollout.yaml

@@ -1656,6 +1656,103 @@ artifacts:
       - type: depends-on
         target: FEAT-FALCON-v0.19.1
 
+  - id: FEAT-FALCON-v0.19.3
+    type: feature
+    title: "v0.19.3 — first PASS verdict under real gz physics (cascade flies)"
+    status: approved
+    description: >
+      LANDED. The closure of v0.19.1 + v0.19.2's two open findings.
+      First time the full falcon-cascade bridge produces a PASS verdict
+      under real Gazebo Harmonic physics — falcon flies under real
+      physics, not just architecturally readied.
+
+      v0.19.3 bench (10-second hover, 70 % constant PWM, open loop):
+        verdict: backend=gazebo steps=1000 climb=385.25 m  wall=11.62s
+        counters: imu_recv=2369 navsat_recv=117 motor_send=1000
+        **PASS**
+
+      Both v0.19.2 sub-issues closed:
+
+      **Finding 1b — SDF motor plugin tuning** (derived against PX4
+      x500's known-flying SDF; gz CLI positive control confirms):
+        - <collision> on every rotor link. ODE treats links without
+          collision as visual-only; thrust didn't transfer through
+          the joint. v0.19.0–v0.19.2 had this gap.
+        - Joint axis: explicit <limit> + spring style; no
+          <damping>. x500's pattern.
+        - commandSubTopic = command/motor_speed (PX4 standard),
+          replacing v0.19.0's cmd_vel.
+
+      **Finding 1a — bridge publish-partition mismatch**:
+        Publisher::publish was called with empty partition,
+        producing FQN @/@/<topic>. gz-sim subscribes on the node's
+        default partition (GZ_PARTITION env or hostname:username),
+        FQN @hostname:user@/<topic>. Different FQNs, no delivery.
+        Fix: publisher.publish(&node.partition(), &msg). One line,
+        explains v0.19.2's `gz topic -i` invisibility too.
+
+      What v0.19.3 proves:
+        - Bridge subscribes IMU + NavSat at configured rates
+          (203 Hz / 10 Hz exactly).
+        - Bridge publishes Actuators with correct partition →
+          MulticopterMotorModel receives + applies torque → thrust
+          → body lift end-to-end.
+        - 385 m climb in 10 s at 70 % PWM = open-loop ballistic
+          ascent; PASS criterion was climb > 0.1 m, well crossed.
+
+      Ships:
+        - examples/falcon-sitl-gz/worlds/falcon-quad.sdf — fully
+          revised against x500.
+        - examples/falcon-sitl-gz/src/physics.rs — partition fix,
+          one line where it matters, plus topic name update to
+          `command/motor_speed`.
+        - bench-evidence/gz-sim/1779860110-gazebo-hover-{harness.log,
+          ticks.csv} + 2026-05-27-v0.19.3-first-pass-verdict.md.
+        - FV-FALCON-SIM-009 + this rollout entry.
+
+      Honestly NOT claimed:
+        - That this is steady hover. 385 m ballistic climb at constant
+          70 % PWM is open loop, no cascade control. Closed-loop
+          hover via the real cascade lands as v0.19.4 (`--scenario=
+          hover` swaps constant PWM for relay-ekf → relay-pos →
+          relay-att → relay-rate → relay-mix-quad).
+        - That the SDF is flight-tuned. motorConstant 8.55e-6 gives
+          5× headroom — fine for flies-vs-doesn't, needs tuning for
+          steady disturbance rejection.
+
+      v0.19.4 candidates:
+        - Closed-loop hover (cascade in the loop; PASS = ±0.5 m for
+          30 s).
+        - Step response (settling < 2 s, overshoot < 20 %).
+        - Mission + geofence-RTL.
+        - Disturbance recovery (gz wind plugin).
+        Each = its own bench-evidence run + rollout entry.
+
+      Verification:
+        - cargo test --workspace --all-targets → 405 passing.
+        - cargo test -p falcon-sitl-gz --features gazebo → 11/11.
+        - cargo build --features gazebo → green.
+        - rivet validate → PASS.
+        - gz sim + bridge bench → PASS, climb=385.25 m.
+
+      Headline: 4 PRs in, the PulseEngine claim "falcon flies under
+      real physics" has positive observational evidence, not just
+      architectural readiness. The trajectory from here is tuning,
+      not unblocking.
+    tags: [falcon, milestone, v0.19.3, gazebo, bench-evidence, pass, first-flight, landed]
+    fields:
+      release-target: "first PASS verdict under real gz physics"
+      bench-date: "2026-05-27"
+      gz-version: "8.11.0"
+      counters-observed: "imu_recv=2369 navsat_recv=117 motor_send=1000"
+      climb_m: 385.247
+      verdict: PASS
+      findings-closed: ["Finding 1a (partition)", "Finding 1b (SDF rotor + joint + topic)"]
+      reference-sdf: "PX4-Autopilot/Tools/simulation/gz/models/x500_base/model.sdf"
+    links:
+      - type: depends-on
+        target: FEAT-FALCON-v0.19.2
+
   - id: FEAT-FALCON-v1.0
     type: feature
     title: "v1.0 — six-domain credit dossier + airframe variants"
@@ -1692,4 +1789,4 @@ artifacts:
       - type: implements
         target: SYSREQ-FALCON-010
       - type: depends-on
-        target: FEAT-FALCON-v0.19.2
+        target: FEAT-FALCON-v0.19.3

artifacts/verification/FV-FALCON-SIM-009.yaml

@@ -0,0 +1,104 @@
+artifacts:
+  - id: FV-FALCON-SIM-009
+    type: sw-verification
+    title: "v0.19.3 — first PASS verdict under real gz physics (Findings 1a + 1b closed)"
+    status: approved
+    description: >
+      v0.19.3 closes both v0.19.2 sub-issues simultaneously and lands
+      the first PASS verdict of the full falcon-cascade bridge against
+      real Gazebo Harmonic physics.
+
+      v0.19.3 bench (10-second hover, constant 70 % PWM):
+        verdict: backend=gazebo steps=1000 climb=385.25 m  wall=11.62s
+        counters: imu_recv=2369 navsat_recv=117 motor_send=1000
+        PASS
+
+      Both findings closed:
+
+      **Finding 1b — SDF motor plugin tuning** (3 corrections vs v0.19.0
+      world, derived from PX4 x500's known-flying SDF):
+        - `<collision>` element on every rotor link. ODE physics
+          treats a link without `<collision>` as visual-only; motor
+          thrust applied to such a link doesn't transfer to the
+          parent body via the revolute joint. v0.19.0–v0.19.2 had this
+          gap → 0 m climb on the v0.19.2 bench.
+        - Joint axis: explicit `<limit><lower>-1e+16</lower>
+          <upper>1e+16</upper></limit>` +
+          `<spring_reference>0</spring_reference>
+          <spring_stiffness>0</spring_stiffness>` replacing v0.19.2's
+          `<dynamics><damping>0.001</damping>`. x500's pattern;
+          lets the rotor joint spin freely under motor torque.
+        - `<commandSubTopic>command/motor_speed</commandSubTopic>`
+          (PX4 standard) replacing v0.19.0's `cmd_vel`. Topic is
+          `/<model>/command/motor_speed`.
+
+        Positive control: `gz topic -t /quad/command/motor_speed
+        -m gz.msgs.Actuators -p "velocity: [1000, 1000, 1000,
+        1000]"` lifted body from z=0.02 m to z=187 m in 3 s.
+
+      **Finding 1a — bridge publish-direction discovery**:
+        After Finding 1b's SDF fix, the bridge still showed
+        motor_send=1000 / climb=0.02 m. Root cause:
+        `Publisher::publish(partition, msg)` constructs the topic
+        FQN as `@<partition>@/<topic>`. The bridge was passing `""`
+        for partition (v0.18 shortcut), producing FQN
+        `@/@/quad/command/motor_speed`. gz-sim's
+        MulticopterMotorModel subscribed via gz-transport-cpp on the
+        node's effective partition (`GZ_PARTITION` env or
+        `hostname:username` default) — FQN
+        `@hostname:user@/quad/command/motor_speed`. Different FQNs,
+        no delivery.
+
+        Fix:
+        ```rust
+        let node_partition = node.partition();   // gz default
+        ...
+        publisher.publish(&node_partition, &msg)
+        ```
+        (replacing `publisher.publish("", &msg)`).
+
+      Also explains why `gz topic -i` showed our bridge un-registered
+      as a publisher despite `advertise()` returning Ok: gz CLI's
+      `gz topic -i` filters by the same default partition, so a
+      publisher under an empty/different partition is invisible.
+
+      What v0.19.3 proves:
+        - falcon-cascade bridge subscribes IMU + NavSat at configured
+          rates (203 Hz / 10 Hz; 2369 + 117 frames in 11.6 s).
+        - Bridge publishes per-tick gz.msgs.Actuators to
+          /<model>/command/motor_speed with the correct partition.
+        - Plugins receive + act on the messages → torque → rotor
+          velocity → thrust → body lift, end-to-end.
+        - 385 m climb under 70 % PWM in 10 s = ballistic ascent
+          (open loop, no closed-loop control). PASS criterion was
+          climb > 0.1 m → flight confirmed.
+
+      Honestly NOT claimed:
+        - That this is steady hover. 385 m ballistic climb is open
+          loop. Closed-loop hover (cascade) → v0.19.4 scenario.
+        - That the SDF is flight-tuned. motorConstant gives 5×
+          headroom — good for "flies vs doesn't fly", not for
+          steady disturbance rejection.
+
+      Tests:
+        - cargo test --workspace --all-targets → 405 passing.
+        - cargo test -p falcon-sitl-gz --features gazebo → 11/11.
+        - cargo build --features gazebo → green.
+        - rivet validate → PASS.
+        - gz sim + bridge bench → PASS (this run).
+    tags: [falcon, sim, gazebo, bench-evidence, pass, first-flight, v0.19.3]
+    fields:
+      bench-evidence-dir: bench-evidence/gz-sim/
+      bench-date: "2026-05-27"
+      gz-version: "8.11.0"
+      counters-observed:
+        imu_recv: 2369
+        navsat_recv: 117
+        motor_send: 1000
+      climb_m: 385.247
+      verdict: PASS
+      findings-closed: ["Finding 1a (publish partition)", "Finding 1b (SDF rotor + joint + topic)"]
+      reference-sdf: "PX4-Autopilot/Tools/simulation/gz/models/x500_base/model.sdf"
+    links:
+      - type: verifies
+        target: SWREQ-FALCON-SIM-P04

bench-evidence/gz-sim/1779860110-gazebo-hover-harness.log

@@ -0,0 +1,14 @@
+falcon-sitl-gz bench-evidence
+backend: gazebo
+scenario: hover
+timestamp: 1779860110
+
+steps:      1000
+net_climb:  385.247 m
+min_alt:    -0.000 m
+max_alt:    385.247 m
+wall:       11.624 s
+imu_recv:   2369
+navsat_recv:117
+motor_send: 1000
+verdict:    PASS