AFIO: Open-source ROS 2/PX4 observer detects Offboard degradation from AI inference lag
New observer catches stale setpoints and jitter before your drone crashes.
AFIO, released by developer zc_Liu, is a zero-intrusive ROS 2/PX4 observer designed to expose a critical failure surface in autonomous drones: when a companion computer running VIO, neural planning, or heavy perception workloads causes PX4 to receive stale or jittery Offboard intent. Unlike failsafe logic, AFIO is observe-only—it doesn't intercept control or publish recovery commands. It subscribes to trajectory setpoint, offboard control mode, and vehicle odometry topics, then publishes standard ROS diagnostics including setpointAgeMs, setpointJitterMs, and dominantCause (e.g., SETPOINT_JITTER, STALE_STREAM).
The project includes an ai_latency_injector_node for controlled testing. In a Gazebo/PX4 SITL circular tracking mission, AFIO detected no degradation at 0-80ms of injected AI lag, but at 150ms it emitted a WARN with SETPOINT_JITTER (p95 setpointJitter: 119.0ms, flightResidual: 0.483). At 300ms, the system hit ERROR with STALE_STREAM and RESYNCING (p95 jitter: 277.2ms, residual: 1.107). This allows developers to regression-test their Offboard stacks under realistic compute loads. AFIO is licensed Apache-2.0 and targets ROS 2 Humble/Jazzy with PX4 v1.14/v1.15.
- AFIO passively monitors PX4 Offboard topics (setpoint, control mode, odometry) and outputs diagnostics like setpointAgeMs and setpointJitterMs.
- Controlled AI-lag injection tests show degradation begins at 150ms latency, with SETPOINT_JITTER warnings appearing before hard failure.
- Supports detection of DDS jitter, companion-compute overload, and thermal throttling on Jetson/RK/NPU boards without modifying the flight stack.
Why It Matters
Early detection of AI-induced flight degradation prevents crashes in autonomous drones, improving safety for commercial and research UAV operations.