Abstract
Discrete control systems, as considered here, refer to the control theory of discrete-time Lagrangian or Hamiltonian systems. These discrete-time models are based on a discrete variational principle and are part of the broader field of geometric integration. Geometric integrators are numerical integration methods that preserve geometric properties of continuous systems, such as conservation of the symplectic form, momentum, and energy. They also guarantee that the discrete flow remains on the manifold on which the continuous system evolves, an important property in the case of rigid-body dynamics. This paper advocates for a "discrete-first" approach to control theory, building controllers directly on top of geometric integrators rather than approximating continuous-time controllers.
Introduction
In nonlinear control, theorists rely on differential geometric and dynamical systems techniques to prove properties such as stability, controllability, and optimality. However, a significant gap exists between high-level geometric control theory and its numerical execution. Continuous-time theories are typically implemented using standard numerical solvers like Runge-Kutta, which are "geometry-blind." These solvers introduce artificial numerical dissipation or drift, causing simulated systems to lose energy or drift off their intended manifolds (e.g., rotation groups). This work addresses this deficiency by restricting the approximation to the choice of a discrete-time model and developing an associated control theory that preserves the crucial underlying geometric structure.
Methods
The proposed framework utilizes geometric integration to ensure that the discrete-time model possesses its own exact conservation laws. The approach is built on several pillars:
- Variational Integrators: Instead of discretizing differential equations of motion, the framework discretizes Hamilton’s Principle. By finding the stationary point of a discrete "action sum," the resulting Discrete Euler-Lagrange equations automatically preserve symplectic structure and momentum maps.
- Lie Group Methods: For systems involving rotations, discrete updates are defined as group operations (e.g., $g_{k+1} = g_k f_k$), ensuring the system remains on the Lie group manifold (like $SO(3)$) to machine precision without needing messy local coordinates or re-projection.
- Discrete Mechanics and Optimal Control (DMOC): Discrete equations of motion are treated as hard constraints in optimization problems, allowing for larger time steps and lower computational loads.
Results
The methods were evaluated using benchmarks including mass-spring systems, planar and spherical pendulums, and dumbbell rigid bodies.
Energy and Manifold Conservation
Variational integrators (VI) demonstrated near-zero energy drift (mean variation $\approx 10^{-13}$), whereas standard 4th-order Runge-Kutta (RK45) methods showed significant dissipation over time. For the spherical pendulum, the VI preserved the unit-length constraint of the position vector to machine precision ($10^{-15}$), while RK45 drifted by $10^{-2}$.
Computational Efficiency
In optimal control tasks, the DMOC approach found accurate solutions with much larger time steps than traditional methods, significantly reducing CPU time for complex maneuvers like spacecraft attitude changes. The discrete methods also proved less sensitive to numerical ill-conditioning.
Conclusion
The "discrete-first" philosophy views geometric integrators as intrinsically discrete dynamical systems with their own conservation laws. Future work will focus on extending these methods to feedback control via Discrete Receding Horizon Control and developing "Geometric Filtering" for state estimation on non-Euclidean manifolds. Additionally, exploring "Forced Symplectic-Energy-Momentum" integrators may further refine energy conservation in the presence of external forces.
Summary updated