Data-Driven Estimation of Quadrotor Motor Efficiency via Residual Minimization

Basic Information

• Paper ID: 2510.11388
• Title: Data-Driven Estimation of Quadrotor Motor Efficiency via Residual Minimization
• Authors: Sheng-Wen Cheng (University of Texas at Austin), Teng-Hu Cheng (National Yang Ming Chiao Tung University)
• Classification: eess.SY cs.SY (Systems and Control)
• Publication Date: October 13, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.11388

Abstract

This paper proposes a data-driven framework based on residual minimization for online estimation of quadrotor motor efficiency. The problem is formulated as a constrained nonlinear optimization problem, solved by minimizing trajectory residuals between measured flight data and quadrotor dynamics model predictions. An online estimation scheme is implemented using a sliding window strategy, with iterative reweighted least squares (IRLS) combined with primal-dual interior point methods to efficiently solve the optimization problem, enforcing inequality constraints through logarithmic barrier functions. Robust z-score weighting based on median absolute deviation (MAD) is employed to reject outliers, demonstrating particular effectiveness in motor saturation scenarios with smaller estimation spikes compared to the EKF baseline method.

Research Background and Motivation

Problem Definition

Quadrotors depend critically on rapid motor response to achieve stable and precise flight. The performance of the propulsion system directly impacts energy efficiency, flight duration, and maneuverability. Motor efficiency, a key but typically unmeasurable factor, degrades due to elevated temperatures, component aging, mechanical wear, and battery voltage fluctuations.

Research Significance

Effective motor efficiency estimation is valuable for:

1. Fault Detection and Isolation (FDI): Early detection of motor performance degradation
2. Health Monitoring: Real-time system state assessment
3. Predictive Maintenance: Prevention of in-flight failure risks
4. Energy Efficiency Optimization: Extended flight endurance

Limitations of Existing Methods

Existing research primarily focuses on mass, inertia, or center-of-mass parameter identification and actuator fault detection, with relatively limited attention to motor efficiency estimation. Traditional filtering methods (e.g., EKF, UKF) tend to produce estimation spikes when handling abrupt changes and outliers, lacking robustness.

Core Contributions

1. Innovative Optimization Framework: Proposes a motor efficiency estimation method based on constrained nonlinear optimization, achieving parameter identification through trajectory residual minimization
2. Efficient Solution Algorithm: Combines IRLS and primal-dual interior point methods, providing computationally efficient and robust solutions
3. Outlier Handling Mechanism: Employs robust z-score weighting based on MAD, effectively suppressing the impact of anomalous measurements
4. Real-time Application Capability: The sliding window strategy enables onboard real-time implementation

Methodology Details

Task Definition

Input: Measured quadrotor states (position x, velocity v, angular velocity Ω, rotation matrix R) and control inputs (collective thrust fc, torque M) Output: Efficiency factors for four motors η = η₁, η₂, η₃, η₄ᵀ Constraints: ηₘᵢₙ ≤ ηᵢ ≤ ηₘₐₓ, typically 0 ≤ η ≤ 1

Model Architecture

1. Quadrotor Dynamics Model

Rigid body dynamics describe quadrotor motion:

Translational Dynamics:

ẋ = v                                    (1)
mv̇ = mge₃ - fcRe₃                       (2)

Rotational Dynamics:

Ṙ = R[Ω]×                               (3)
JΩ̇ + Ω× JΩ = M                          (4)

2. Thrust Allocation and Efficiency Modeling

The relationship between motor thrust and collective force and torque:

[fc,actual; M1,actual; M2,actual; M3,actual] = ΛEΛ⁻¹[fc; M1; M2; M3]  (14)

where Λ is the thrust allocation matrix and E = diag(η₁, η₂, η₃, η₄) is the efficiency matrix.

3. Trajectory Residual Definition

Four types of residuals are defined:

• Velocity Residual: rv,t(s) = vt - v̂t(s)
• Position Residual: rx,t(s) = xt - x̂t(s)
• Angular Velocity Residual: rΩ,t(s) = Ωt - Ω̂t(s)
• Attitude Residual: rR,t(s) = ½trI - δRᵀδR̂(s)

Technical Innovations

1. Constrained Optimization Formulation

The estimation problem is formulated as:

min F(st) = ½‖r(st)‖²G + γ/2‖st - st-1‖²
s.t. ηmin ≤ ηi ≤ ηmax, i = 1,···,4

The objective function comprises trajectory residual terms and temporal smoothing regularization.

2. Primal-Dual Interior Point Method

Logarithmic barrier functions handle inequality constraints:

min F(st) - 1/β Σlog(-φi(st))

Solved iteratively via KKT conditions and Newton's method.

3. Robust Outlier Handling

Z-score weighting based on median absolute deviation (MAD):

zi = |ei - m̃|/max(MAD, εmin)
wi = max(1/(1 + (zi/zsoft)^p), wmin)

Combines soft attenuation and hard rejection strategies to enhance robustness.

Experimental Setup

Simulation Platform

Implemented in MATLAB, simulating F450 quadrotor platform characteristics:

• Mass: 1.0 kg
• Inertia Matrix: diag(0.01466, 0.01466, 0.02848) kg·m²
• Motor Arm Length: 0.225 m

Test Scenarios

1. Voltage-Induced Progressive Degradation: Motor efficiency decays exponentially with battery voltage
2. Abrupt Fault Injection: Motor efficiency drops to 0.5 during specific time intervals
3. Random Thrust Noise: Thrust noise with σf = 0.07 is added

Comparison Methods

Extended Kalman Filter (EKF) serves as the baseline method, modeling motor efficiency as a random walk process.

Evaluation Metrics

• Root Mean Square Error (RMSE)
• Standard Deviation
• Maximum Estimation Spike

Experimental Results

Main Results

Across various degradation and fault scenarios, the proposed method demonstrates comparable long-term accuracy to the EKF baseline while exhibiting significant robustness advantages during transient periods:

1. Progressive Degradation Scenario: Both methods accurately track gradual efficiency changes
2. Abrupt Fault Scenario: EKF exhibits pronounced spikes, while the proposed method maintains smooth estimates
3. Combined Scenario: The proposed method maintains stable performance in complex scenarios

Robustness Analysis

Figure 6 demonstrates that the proposed method's maximum estimation spikes are significantly smaller than EKF in abrupt fault and combined scenarios:

• Abrupt Fault Scenario: Spike reduction of approximately 60-80%
• Combined Scenario: Spike reduction of approximately 50-70%

Convergence Performance

• Estimation Convergence: Starting from initial guess of 0.5, converges to true value within one complete interior point method cycle
• Optimization Convergence: Primal residual, dual residual, and surrogate duality gap decrease monotonically, satisfying KKT optimality conditions

Related Work

Quadrotor System Identification

Existing research primarily addresses:

• Parameter Identification: Mass, inertia, and center-of-mass estimation 1-6
• Fault Detection: Neural networks 8, augmented state Kalman filtering 9, sparse identification 10
• Learning Methods: Diffusion model learning for unmodeled effects 7

Motor Efficiency Estimation

Limited related research exists; the authors' prior work 13 proposed a UKF-based motor efficiency estimator for multi-rotor cooperative systems.

Conclusions and Discussion

Main Conclusions

1. Effectiveness: The proposed framework accurately estimates quadrotor motor efficiency
2. Robustness: Demonstrates stronger disturbance rejection in abrupt change scenarios compared to traditional filtering methods
3. Real-time Capability: The sliding window strategy supports online implementation

Limitations

1. Computational Complexity: Optimization solving requires greater computational resources than simple filtering methods
2. Parameter Tuning: Requires appropriate setting of window length, weight parameters, etc.
3. Unmodeled Effects: Does not account for wind disturbances and external perturbations

Future Directions

1. Learning Enhancement: Integration of learning methods to handle unmodeled dynamics
2. Distributed Implementation: Distributed estimation for multi-agent cooperative scenarios
3. Hardware Validation: Experimental verification on actual flight platforms

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Strict mathematical modeling based on optimization theory
2. Methodological Innovation: Transforms system identification into constrained optimization, providing novel solution approaches
3. Practical Value: Directly addresses practical application requirements with engineering value
4. Robust Design: Well-designed outlier handling mechanism enhances practical reliability

Weaknesses

1. Simulation-Only Validation: Lacks verification through actual flight experiments
2. Scenario Limitations: Primarily considers motor efficiency degradation without addressing other fault types
3. Computational Efficiency: While claiming suitability for onboard implementation, lacks specific computational time analysis
4. Parameter Sensitivity: Insufficient discussion of algorithm parameter effects on performance

Impact and Significance

1. Academic Contribution: Provides new optimization perspectives for UAV system identification
2. Application Prospects: Shows good potential for UAV health monitoring and fault diagnosis applications
3. Scalability: Framework exhibits good extensibility applicable to other parameter estimation problems

Applicable Scenarios

1. Commercial UAVs: Commercial quadrotor platforms requiring high reliability
2. Long-Endurance Missions: Extended flight tasks with high energy efficiency requirements
3. Autonomous Maintenance: Intelligent UAV systems supporting predictive maintenance
4. Multi-Agent Cooperation: Health monitoring for large-scale UAV swarms

References

The paper cites 25 relevant references, primarily focused on quadrotor modeling and control, system identification, and optimization algorithms, providing solid theoretical foundations for the research.

Overall Assessment: This is a technically sound paper proposing an innovative motor efficiency estimation method with contributions in both theoretical modeling and algorithm design. Although currently validated only through simulation, the method demonstrates promising practical prospects. Subsequent work should prioritize actual flight verification and computational efficiency optimization.