System Identification

Large-domain modeling, system identification, and flight testing of UAVs

Overview

Accurate flight dynamic models are essential for model-based control, estimation, and autonomy. For UAVs, however, obtaining nonlinear models valid across a wide range of flight conditions is challenging. Traditional approaches often capture only small perturbations around a nominal condition, weakening the guarantees of any controller or estimator designed from them. Our research develops nonlinear multirotor, fixed-wing, and vertical-takeoff-and-landing (VTOL) models that balance accuracy and practicality, along with system identification methods that remain safe even for inherently unstable aircraft.

Approach

System identification for UAVs requires overcoming several challenges: nonlinear aerodynamics, instability, and the need for safe automated experiments. Our approaches have addressed these challenges through three main directions:

1. Nonlinear multirotor and VTOL modeling
   • Models derived from blade-element and momentum theory, valid across diverse flight conditions
   • Physics-informed simplifications enable tractable estimation and control design
2. Safe system identification for unstable aircraft
   • Framework leverages stability guarantees from a robust LPV H₂/H_∞ controller
   • Controller executes specially designed reference signals that decorrelate model regressors, ensuring accurate parameter estimation
   • Enables rich input/output data collection without risking instability
3. Spin and stall dynamics modeling
   • Data-driven aerodynamic models for fixed-wing aircraft in a stall-spin regime
   • Supports control and estimation strategies in these extreme flight conditions

Why It Matters

• Safety assurance: Stabilizes inherently unstable UAVs while performing aggressive excitation for system identification.
• Scalable methods: Applicable to both small UAVs and larger, high-cost vehicles where excitation under non-robust control is high-risk.
• Advanced autonomy: Provides the modeling foundation needed for weather-tolerant air mobility operations.

Selected Publications

• Practical Nonlinear Flight Dynamic Modeling for Multirotor Aircraft — Develops compact nonlinear multirotor models valid across diverse conditions, balancing fidelity and identifiability.
• Robust Linear Parameter-Varying Control for Safe and Effective Unstable Aircraft System Identification — Introduces an LPV H₂/H∞ framework that stabilizes UAVs during excitation and executes decorrelating trajectories for accurate large-domain model identification.
• Development and Evaluation of Multirotor Flight Dynamic Models for Estimation and Control — Presents nonlinear multirotor models derived from blade-element and momentum theory, validated through simulation and experiments.
• Spin Aerodynamic Modeling for a Fixed-Wing Aircraft Using Flight Data — Identifies a nonlinear aerodynamic model for stall-spin regimes using flight test data, outperforming nominal models in extreme flight.
• Remote Uncorrelated Pilot Input Excitation Assessment for Unmanned Aircraft Aerodynamic Modeling — Proposes techniques for ensuring pilot inputs yield uncorrelated data suitable for aerodynamic model identification.
• Flight Test Approach for Modeling and Control Law Validation for Unmanned Aircraft — Demonstrates a progressive build-up approach to flight testing, combining system ID with nonlinear control validation.
• Spin Aerodynamic Modeling for a Fixed-Wing Aircraft Using Flight Data — Early results on spin aerodynamic modeling with novel excitation signal design.