A Narwhal-Inspired Sensing-to-Control Framework for Small Fixed-Wing Aircraft

TL;DR

This work tackles the challenge of achieving agile, robust fixed-wing UAV control under gusts by integrating a narwhal-inspired upstream sensing boom with a physics-informed, control-affine learning model and a convex control allocator. The method calibrates distant Pitot probes to estimate $V_a$, $\alpha$, and $\beta$, learns a dynamic model $y=A_\phi(o)+B_\phi(o)u$ that respects symmetry, and solves a regularized least-squares problem to realize desired wrenches with smooth actuation. Experiments show that wing-pressure sensing reduces force-estimation error by roughly 25–30%, improves generalization under distribution shift (~12% degradation vs ~44% for baselines), and achieves notable tracking gains in normal force, validating the approach's potential for safer, more capable fixed-wing autonomy. The approach offers a practical pathway to combine bio-inspired sensing with structure-aware learning for robust, efficient, and agile UAV operation in challenging environments.

Abstract

Fixed-wing unmanned aerial vehicles (UAVs) offer endurance and efficiency but lack low-speed agility due to highly coupled dynamics. We present an end-to-end sensing-to-control pipeline that combines bio-inspired hardware, physics-informed dynamics learning, and convex control allocation. Measuring airflow on a small airframe is difficult because near-body aerodynamics, propeller slipstream, control-surface actuation, and ambient gusts distort pressure signals. Inspired by the narwhal's protruding tusk, we mount in-house multi-hole probes far upstream and complement them with sparse, carefully placed wing pressure sensors for local flow measurement. A data-driven calibration maps probe pressures to airspeed and flow angles. We then learn a control-affine dynamics model using the estimated airspeed/angles and sparse sensors. A soft left/right symmetry regularizer improves identifiability under partial observability and limits confounding between wing pressures and flaperon inputs. Desired wrenches (forces and moments) are realized by a regularized least-squares allocator that yields smooth, trimmed actuation. Wind-tunnel studies across a wide operating range show that adding wing pressures reduces force-estimation error by 25-30%, the proposed model degrades less under distribution shift (about 12% versus 44% for an unstructured baseline), and force tracking improves with smoother inputs, including a 27% reduction in normal-force RMSE versus a plain affine model and 34% versus an unstructured baseline.

Figures (6)

• Figure 1: Fixed-wing vehicles operate across highly diverse terrain and environments, such as (A) through populated urban environments, (B) across vast deserts, or (C) trailing behind other aerial vehicles in formation flight. The underlying challenging gust are vortex dynamics which are highly time-varying. (D) To stay ahead of the "curve", we built in-house developed, avionic sensors that measures incoming gust (in loose language, "dirty" flow) before it arrives at the wing, and affect drone's dynamics. This approach is inspired by the Narwhal whale, where its tooth protrudes forward, and detects sea water salinity and potentially pressure signals nweeia2014.
• Figure 2: The model fixed-wing drone, model positioning system, and the gust generator in a wind tunnel. (A) The fixed-wing are instrumented with in-house developed avionics as well as an onboard Nvidia Jetson Nano Orin computer. (B) The test environment in the wind tunnel includes a gust generator as well as an in-house developed two-axis rotation bed (or model position system).
• Figure 3: Sensor-to-control pipeline.Section \ref{['subsec:alg_pitot']}: Dual multi-hole Pitot and sparse wing-pressure measurements are calibrated to estimate the flow state $(V_a,\alpha,\beta)$. Section \ref{['subsec:alg_learning']}: A symmetry-aware, control-affine model maps onboard observations to aerodynamic forces and moments. Section \ref{['subsec:opt']}: A convex controller then computes surface deflections to realize a desired wrench.
• Figure 4: Estimated vs. ground-truth wind features. Columns: Pitot tube 0 (left) and Pitot tube 1 (right). Rows: airspeed $V_a$, angle of attack $\alpha$, and sideslip angle $\beta$. The gust is positioned near probe 0.
• Figure 5: $B_\phi^\top$ heatmaps. Left: model trained with symmetry regularization; right: without symmetry regularization. The regularized map is more aerodynamically plausible.