Model-Reference Adaptive Flight Control of the 95-mg Bee++

Technical Summary: Model-Reference Adaptive Flight Control of the 95-mg Bee++

Problem Statement
The paper addresses the challenge of achieving high-performance positional tracking for the Bee++, a 95-mg insect-scale flapping-wing aerial vehicle. Due to the robot's microscopic scale and complex fabrication, its dynamic model suffers from significant parameter uncertainty, particularly regarding the inertia matrix and actuator mapping. Furthermore, the Bee++'s low inertia makes it highly susceptible to stochastic and systematic disturbances during flight, including tether tension, wind gusts, and aerodynamic forces arising from wing misalignment. These factors complicate the execution of high-speed maneuvers and precise hovering, necessitating a control architecture capable of adapting to model uncertainties and external disturbances in real-time.

Methodology
The authors propose a Model-Reference Adaptive Control (MRAC) framework designed for rigid-body flapping-wing robots operating in 3D space. The core methodology involves the following components:

• Dynamic Modeling: The translational dynamics are modeled using Newton's second law, incorporating an unknown aggregated disturbance term $d(t)$. This disturbance is parameterized as a linear combination of Gaussian-based radial basis functions (RBFs), $d(t) = W^T \phi(x)$, where $W$ is a matrix of unknown parameters and $x$ represents the state vector (position and velocity).
• Control Law: A stabilizing control law is derived that includes a standard PID-like structure (proportional, integral, and derivative terms) combined with a feedforward term and an adaptive term, $f_a = \hat{W}^T \phi(x)$. The adaptive term is specifically designed to cancel the unknown disturbance.
• Adaptation Law: An adaptive law is derived to update the estimate $\hat{W}$ in real-time. This law is constructed using a Lyapunov stability analysis. By defining a Lyapunov function candidate and ensuring its time derivative is negative semi-definite, the authors derive an update rule ($\dot{\hat{W}} = -\gamma \phi \tilde{z}^T P^T B$) that guarantees the tracking error converges to zero.
• Stability Analysis: The authors prove that the closed-loop system signals remain bounded and that the tracking error asymptotically converges to zero. The analysis demonstrates that the adaptive method effectively rejects disturbances by forcing the system's state to track a zero-disturbance reference model.

Key Contributions

• MRAC Architecture for Micro-Robots: The paper introduces a specific MRAC architecture tailored for the unique constraints of insect-scale flapping-wing vehicles, addressing both model uncertainty and external disturbances.
• Rigorous Stability Proof: The work provides a formal stability analysis using Lyapunov theory, proving the uniform stability of the equilibrium and the asymptotic convergence of the tracking error.
• Real-Time Experimental Validation: The approach is validated through real-time flight experiments on the physical Bee++ robot, demonstrating its practical applicability rather than relying solely on simulation.

Experimental Results
The proposed MRAC scheme was evaluated against a benchmark non-adaptive controller (from prior work [1]) through five back-to-back hovering experiments with a constant reference position. The results indicate:

• Superior Performance: While both controllers could stabilize the robot, the MRAC scheme significantly outperformed the non-adaptive benchmark.
• Quantitative Improvements: The MRAC controller reduced the mean Root-Mean-Square (RMS) position errors by 15% in the $n_1$ direction, 44% in the $n_2$ direction, and 35% in the $n_3$ direction compared to the benchmark.
• Specific Error Metrics: For the MRAC scheme, the mean RMS errors were $2.32 \pm 0.60$ cm ($n_1$), $0.87 \pm 0.39$ cm ($n_2$), and $0.13 \pm 0.03$ cm ($n_3$).
• Adaptive Behavior: The time evolution of the adaptive force term ($f_a$) demonstrated the controller's active compensation for disturbances during the flight.

Significance and Claims
The paper positions this work as a critical step toward enabling high-speed maneuvers and advanced attitude control for insect-scale robots. The authors claim that the proposed MRAC framework successfully allows the Bee++ to adapt its behavior to counteract disturbances and model uncertainty, thereby enabling high-performance positional tracking. The study demonstrates that the approach is suitable, functional, and capable of high performance in real-world flight conditions, as evidenced by the experimental data. The work does not claim to solve all control challenges for micro-robots but establishes a robust foundation for adaptive control in the presence of significant uncertainty and disturbance.