Physics-Informed Neural Networks Solve Nonlinear Output Regulation
A new approach utilizing a physics-informed neural network (PINN) has been developed by researchers to tackle the full-information output regulation challenge for nonlinear systems. This technique approximates the zero-regulation-error manifold and feedforward input by minimizing residuals while adhering to boundary and feasibility conditions, eliminating the need for precomputed trajectories or labeled data. The operator learned through this method effectively translates exosystem states into steady-state plant states and inputs, facilitating real-time inference and generalization across various exosystem families. This research, which addresses the regulator equations—a set of PDEs with an algebraic constraint—has been published on arXiv under the identifier 2511.13595.
Key facts
- Addresses full-information output regulation for nonlinear systems
- Assumes states of plant and exosystem are known
- Perfect tracking/rejection via invariant manifold and feedforward input
- Regulator equations are PDEs with algebraic constraint
- PINN directly approximates π(w) and c(w)
- No precomputed trajectories or labeled data needed
- Enables real-time inference and generalization
- Published on arXiv with ID 2511.13595
Entities
Institutions
- arXiv