Robust Adaptive Geometric Tracking Controls on SO(3) with an Application to the Attitude Dynamics of a Quadrotor UAV

• Minimal attitude coordinates are local and can become singular in large rotations, while quaternion control can be discontinuous or exhibit unwinding because S3 double-covers SO(3),,,.Source: §I · p. 1
• Geometric control characterizes nonlinear manifolds intrinsically, avoiding singularities and ambiguities associated with local coordinates or improper configuration representations,,.Source: §I · p. 1
• The proposed SO(3) adaptive controller tracks attitude and angular-velocity commands without known inertia, while its robust extension bounds tracking and estimation errors under modeling errors or disturbances.The robust controller also handles unstructured or nonharmonic uncertainties without requiring their frequencies.Source: §I · p. 1

• The model considers a fully actuated rigid body whose orientation is represented by a rotation matrix R in SO(3), the group of 3 × 3 orthogonal matrices with determinant one.Source: §II · p. 1
• The body-frame inertia matrix is J, and the rigid body's angular velocity and control moment are represented by Ω and u in R^3.Source: §II · p. 1
• The dynamics include disturbances attributed to modeling errors or system noise, and the hat map represents cross products through skew-symmetric matrices, with the vee map as its inverse.Source: §II · p. 2

• The control systems are designed to follow a given smooth attitude command Rd(t) in SO(3).Source: §III · p. 2
• The attitude command's desired angular velocity is denoted by Ωd.Source: §III · p. 2

A. Attitude Error Dynamics

• A configuration error function Ψ is a smooth positive-definite measure of the difference between current and desired configurations, from which configuration and velocity error vectors are defined.Control inputs are designed as functions of these errors using Lyapunov analysis on the manifold.Source: A. Attitude Error Dynamics · p. 2
• The cited properties state that Ψ is locally positive definite about R = Rd and that Rd belongs to the critical-point set of Ψ.Source: A. Attitude Error Dynamics · p. 2
• When Ψ is below ψ < h1, the three non-desired critical points are avoided, ensuring eR is nonzero and the rotation-vector norm is less than π.Source: A. Attitude Error Dynamics · pp. 2–3
• The section derives the attitude and angular-velocity error dynamics, including the matrix E(R, Rd) and the command-induced angular acceleration αd.Source: A. Attitude Error Dynamics · p. 3

B. Adaptive Attitude Tracking

• The controller updates an inertia estimate online and uses attitude and angular-velocity feedback with inertia compensation, avoiding the need to know the exact inertia matrix.The section notes that exact inertia measurement is difficult.Source: B. Adaptive Attitude Tracking · p. 3
• With zero disturbance and known inertia eigenvalue extrema, attitude and angular-velocity tracking errors converge asymptotically while tracking and inertia-estimation errors remain stable and uniformly bounded.The Lyapunov analysis and Barbalat’s lemma establish convergence.Source: B. Adaptive Attitude Tracking · pp. 3–4
• Convergence of the attitude error vector to zero does not ensure R→Rd: three additional critical attitudes also have zero error, though they are saddle equilibria with a lower-dimensional union of stable manifolds.The paper describes the resulting property as almost-global stabilization.Source: B. Adaptive Attitude Tracking · p. 4
• The eigenvalue assumption can be relaxed to an upper bound on the minimum eigenvalue and a lower bound on the maximum eigenvalue, which are simpler to estimate.Source: B. Adaptive Attitude Tracking · p. 4

C. Robust Adaptive Attitude Tracking

• The robust adaptive design assumes the disturbance is bounded by a known constant and introduces a control input and inertia-estimate update law.The robust scheme modifies adaptive tracking, which may become unstable under even small disturbances when uncertainty is not considered.Source: C. Robust Adaptive Attitude Tracking · pp. 4–5
• With sufficiently small σ and ϵ, the zero equilibrium of the attitude, angular-velocity, and inertia-estimation errors is uniformly bounded.For initial conditions satisfying V(0) < d2, solutions exponentially converge to a set; the fixed σ-modification trades asymptotic tracking for boundedness.Source: C. Robust Adaptive Attitude Tracking · pp. 5–6
• The boundedness argument confines trajectories to nested Lyapunov sublevel sets, with V decreasing until solutions enter the smaller set.The result applies for initial conditions inside the larger set and requires sufficiently small σ and ϵ.Source: Figure 1 · p. 5, C. Robust Adaptive Attitude Tracking · pp. 5–6
• Replacing the leakage term’s zero reference with a prior inertia estimate can reduce the ultimate bound, while switching modifications may improve convergence at the cost of discontinuities.A time-decaying ϵ is another proposed way to reduce the ultimate bound; the corresponding stability analyses are deferred to future work.Source: C. Robust Adaptive Attitude Tracking · p. 6

IV. NUMERICAL EXAMPLES

• The simulations use a Lie group variational integrator to preserve the geometric structure of the attitude dynamics.The passage notes that general-purpose numerical integrators may fail to preserve SO(3) structure and yield unreliable results for complex rigid-body maneuvers.Source: §IV · p. 6
• Under disturbances, the robust controller brings attitude and angular-velocity tracking errors close to zero and keeps the inertia estimate bounded.The nominal adaptive controller’s errors do not converge to zero asymptotically under disturbances, and its inertia estimate and control inputs fluctuate.Source: §IV · p. 6
• Without disturbances, the adaptive controller tracks the attitude command accurately and its inertia estimate converges to a fixed matrix.The simulation compares this case with nominal adaptive tracking under disturbances and robust adaptive tracking under disturbances.Source: §IV · p. 6

V. EXPERIMENT ON A QUADROTOR UAV

• The hardware system includes a Gumstix Overo computer, Microstrain attitude sensor, Phifun motor controller, and Roxxy brushless motors.The computer runs non-realtime Linux and communicates with a ground station over Wi-Fi.Source: §V · pp. 6–7
• The UAV is attached to a spherical joint, creating destabilizing gravitational moment because its rotation center lies below its center of gravity.An XBee RF module provides an extra safety switch.Source: §V · p. 7
• The robust adaptive controller is augmented with a term to eliminate the gravitational moment in the quadrotor setup.The experiment’s disturbances mainly arise from gravity-cancellation error, spherical-joint friction, sensor noise, and thrust-measurement error.Source: §V · p. 7
• The experiment shows good attitude-command tracking with bounded inertia estimates, while the second attitude-error and third angular-velocity-error components are relatively large.The quadrotor is tested on a spherical joint, where the center of rotation lies below the center of gravity.Source: §V · p. 7

• The control system is constructed directly on SO(3) to avoid singularities and ambiguities inherent to other attitude representations.Source: §VI · p. 7
• Without disturbances, the adaptive system asymptotically follows an attitude command without knowledge of the inertia matrix.Source: §VI · p. 7
• The robust adaptive system is proposed to eliminate disturbance effects.Source: §VI · p. 7
• Numerical examples and a quadrotor UAV hardware experiment illustrate the stated properties.Source: §VI · p. 7