• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Sunday, November 22, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Data-Driven Predictive Control Framework Stabilizes Quadrotors Against Actuator Faults and Disturbances

by
November 20, 2026
in Technology
Reading Time: 5 mins read
0
Data-Driven Predictive Control Framework Stabilizes Quadrotors Against Actuator Faults and Disturbances

Data-Driven Predictive Control Framework Stabilizes Quadrotors Against Actuator Faults and Disturbances

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A control-engineering research team has formulated a data-driven, fault-tolerant model predictive control architecture designed to preserve the flight stability and trajectory-tracking accuracy of quadrotor unmanned aerial vehicles under severe operating anomalies. Quadrotors possess an underactuated mechanical nature characterized by six spatial degrees of freedom governed through only four physical control inputs. This configuration makes them particularly vulnerable to physical component degradation, including rotor efficiency losses, actuator biases, parameter mismatches, and sudden aerodynamic disturbances such as turbulent wind gusts. To overcome the computational and analytical bottlenecks of earlier control formulations, the researchers integrated physics-based dynamic modeling with real-time radial basis function neural network compensation into a dual-loop cascaded framework. Their analytical derivation and numerical simulations demonstrate that the system maintains robust tracking and closed-loop stability across compound failure modes while operating entirely through computationally lightweight convex quadratic programs.

Standard approaches to quadrotor control have long contended with difficult trade-offs between physical accuracy and online computational tractability. Traditional model predictive control relies on mathematical models derived from first principles, where nonlinear formulations demand substantial onboard processor power to solve non-convex nonlinear programming problems in real time. Conversely, purely data-driven control architectures learn dynamics without nominal physics, frequently failing to deliver the formal safety proofs and recursive feasibility guarantees essential for autonomous flight in safety-critical airspaces. To resolve this dilemma, the authors partitioned the quadrotor dynamics across two natural operational timescales. Rotational dynamics operate at a much faster rate than translational motion, allowing the inner attitude loop to execute at a higher sampling frequency than the outer position loop. This structural separation prevents the severe controllability loss and vertex explosion that occur when engineers attempt to synthesize a monolithic full-state parameter-varying model for an entire vehicle.

In the outer translational control loop, the investigators resolved the non-affine coupling between total rotor thrust and vehicle orientation angles by leveraging differential flatness via feedback linearization. This analytical transformation maps the coupled spatial dynamics into three decoupled, linear time-invariant double-integrator subsystems governed by virtual acceleration inputs. A reverse algebraic calculation extracts the required collective thrust alongside reference roll and pitch angles, enforcing bounded attitude angles to strictly avoid geometric singularities where total vertical control authority would vanish. The resulting outer-loop optimization is formulated as a linear time-invariant model predictive controller. Because the underlying state-space dynamics become strictly linear, the outer controller computes optimal virtual accelerations along a finite prediction horizon by solving a convex quadratic program, penalizing trajectory deviation and virtual control effort while satisfying actuator and state limits.

For the inner attitude control loop, the researchers embedded the fast, nonlinear rotational dynamics into a discrete-time quasi-linear parameter-varying representation. Attitude orientation and angular rates encounter intense cross-axis gyroscopic effects and Coriolis couplings that depend directly on roll rate, pitch rate, and the algebraic sum of the individual rotor speeds. Rather than treating these terms as unstructured noise or relying on iterative nonlinear solvers, the dynamics are structured across a convex polytope bounded by parameter vertices. The scheduling vector is tracked through time and reconstructed along the prediction horizon by shifting prior optimal trajectories. Because the state-space system matrix is represented as a convex combination of invariant vertex matrices, the inner attitude predictive controller is also solved as a convex quadratic program, allowing high-frequency execution with minimal processor overhead.

To counteract unknown physical faults and external environmental disturbances, the architecture incorporates two offline-trained radial basis function neural networks that approximate lumped matched uncertainties affecting the system. Rather than attempting to learn the complete baseline dynamics of the quadrotor from scratch, the networks are deployed strictly as feedforward estimators for unknown lumped disturbances. The position network receives an eighteen-dimensional regressor vector capturing current and delayed position states, past inputs, and vehicle orientation angles, while the attitude network processes a fifteen-dimensional regressor vector. Both neural networks feature twenty hidden neurons utilizing Gaussian basis functions. Hidden-layer centers were determined through k-means clustering across regressor datasets generated from simulated step and sinusoidal trajectory maneuvers, widths were assigned through nearest-neighbor heuristics, and output weight matrices were computed via Moore-Penrose pseudoinverse linear least squares.

A critical technical contribution of the study is the integration of imperfect neural network compensation directly into the formal robustness and stability analysis. Rather than assuming idealized, zero-error neural estimation, the authors established that the residual reconstruction error remains bounded within a known compact polytope. By actively compensating for the estimated portion of the lumped uncertainty within the nominal predictive models, the subsequent robust terminal ingredients only need to accommodate the small residual estimation error rather than the full, unmitigated magnitude of worst-case physical faults. This strategy prevents the severe conservatism and feasible-region shrinkage typical of classical robust predictive controllers. Parameter-dependent terminal feedback gains and a common symmetric positive-definite Lyapunov matrix were synthesized simultaneously by transforming the closed-loop terminal dynamics into a set of linear matrix inequalities solved across all polytopic vertices.

The mathematical proofs establish subsystem recursive feasibility and robust input-to-state stability with respect to bounded estimation errors, showing that closed-loop state trajectories are driven into a compact ellipsoidal invariant terminal set. The overall cascaded system stability was subsequently verified using the input-to-state stability small-gain theorem, ensuring that any residual tracking error escaping from the fast inner loop acts merely as an attenuated disturbance within the outer loop. Numerical validation was conducted within a MATLAB and Simulink simulation environment modeling the physical parameters of an AscTec Hummingbird quadrotor, integrating fourth-order Runge-Kutta numerical integration at a fundamental step size of one millisecond. The online convex quadratic programs were executed using the OSQP solver within the CasADi framework, while comparative nonlinear predictive controllers were evaluated using the Ipopt nonlinear solver.

The evaluation examined three distinct operational flight scenarios along an aggressive rising spiral trajectory combined with sinusoidal heading changes. In the first scenario, the baseline model predictive controller was assessed under fifteen percent parameter variations across vehicle mass, arm length, and moments of inertia. The nominal parameter-varying architecture matched the trajectory tracking accuracy of a fully nonlinear predictive controller while executing up to eleven times faster, demonstrating that the convex quadratic formulation provides sufficient inherent robustness against moderate parameter mismatch without incurring high computational penalties. In the second scenario, the quadrotor was subjected to external force and torque disturbances representing turbulent wind conditions. Standalone predictive control experienced significant path deviations, whereas the neural-network-augmented framework achieved accurate uncertainty estimation, reducing maximum position tracking error from 1.13 meters down to 0.26 meters.

In the third scenario, the control scheme was tested against simultaneous extreme operational stresses, combining fifteen percent model mismatch, persistent environmental disturbances, and severe additive and multiplicative actuator degradation representing rotor efficiency loss and motor bias. The uncompensated predictive controller suffered catastrophic trajectory divergence, exhibiting position errors exceeding two meters. In contrast, the radial-basis-function-augmented controller maintained flight stability and tightly tracked the reference profile, achieving a root mean square position error of 0.11 meters compared to 1.06 meters for the baseline model. While the neural network exhibited brief transient oscillations immediately following abrupt fault injections due to estimation delay, these disturbances were rapidly suppressed as the estimators converged, preventing sustained control biases and avoiding actuator saturation.

The authors noted specific operational boundaries and theoretical limitations inherent to the framework. Because the radial basis function networks are trained offline on finite trajectory data, their mathematical validity depends on local interpolation within data-supported regions of the regressor domain rather than reliable global extrapolation. Consequently, formal stability and recursive feasibility guarantees remain conditional upon operating within the pre-established bounded uncertainty polytope where actuator limits preserve sufficient physical control authority. Furthermore, the framework currently lacks validation through physical experimental flight trials on hardware platforms. The research team identified real-world quadrotor flight testing and the theoretical integration of tube-based predictive control formulations for enhanced constraint tightening as the primary avenues for subsequent investigation.

Subject of Research: Fault-tolerant model predictive control for quadrotor unmanned aerial vehicles

Article Title: Data-driven fault-tolerant MPC for quadrotors: a quasi linear parameter varying approach with RBF neural network compensation

Article References: Senoussi, M. A., Boumehraz, M., Puig, V., Glida, H. E., & Bacha, A. (2026). Data-driven fault-tolerant MPC for quadrotors: a quasi linear parameter varying approach with RBF neural network compensation. International Journal of Intelligent Robotics and Applications. https://doi.org/10.1007/s41315-026-00606-y

Image Credits: AI Generated

DOI: 10.1007/s41315-026-00606-y

Keywords: Quadrotor UAVs, Model Predictive Control, Fault-Tolerant Control, Radial Basis Function Neural Networks, Quasi-LPV Modeling, Feedback Linearization, Input-to-State Stability, Actuator Faults, Convex Quadratic Programming, Disturbance Rejection, Robotics Control Systems, AscTec Hummingbird, Unmanned Aerial Vehicles, Linear Matrix Inequalities, Robust Stability, Cascaded Control Architecture

News Source: Denise Maddox. (November 20, 2026). Data-Driven Predictive Control Framework Stabilizes Quadrotors Against Actuator Faults and Disturbances. Scienmag.

Tags: Actuator FaultsAscTec HummingbirdCascaded Control ArchitectureConvex Quadratic Programmingdisturbance rejectionfault-tolerant controlfeedback linearizationinput-to-state stabilityLinear Matrix Inequalitiesmodel predictive controlQuadrotor UAVsQuasi-LPV Modelingradial basis function neural networksRobotics Control SystemsRobust Stabilityunmanned aerial vehicles
Share12Tweet7Share2ShareShareShare1

Related Posts

Radio Tags Alter Flight Kinematics in Neotropical Butterflies Below Standard Weight Limits

Radio Tags Alter Flight Kinematics in Neotropical Butterflies Below Standard Weight Limits

November 22, 2026
Strut-and-Tie Model Accurately Predicts Flexural Capacity of Prefabricated Bridge Wet Joints

Strut-and-Tie Model Accurately Predicts Flexural Capacity of Prefabricated Bridge Wet Joints

November 22, 2026

Precast Concrete Shear Walls with UHPC Boundary Connections Improve Seismic Performance

November 21, 2026

Portable Raman Spectroscopy Screens Microbial Carotenoids for Solar Cell Fabrication

November 21, 2026

POPULAR NEWS

  • DCAF15 and ANKIB1 Form Enzyme Complex Driving Inflammatory NLRP3 Activation

    DCAF15 and ANKIB1 Form Enzyme Complex Driving Inflammatory NLRP3 Activation

    29 shares
    Share 12 Tweet 7
  • Radio Tags Alter Flight Kinematics in Neotropical Butterflies Below Standard Weight Limits

    29 shares
    Share 12 Tweet 7
  • Gene Promoter Variations Are Associated with Lobed-Leaf Formation in Siberian Kale

    29 shares
    Share 12 Tweet 7
  • Screening Human Fat Cells Uncovers PLCXD1 Enzyme as Insulin-Regulated Lipid Storage Driver

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Mutant p53 Reprograms AP-1 Complexes to Drive Cancer Cell Migration

Automated Telephone Cognitive Tests Track Regional Brain Atrophy Across Alzheimer Continuum

Circulating Vesicle Proteins Interfere With CLDN18.2 Cell and Antibody Therapies

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.