• Corpus ID: 221139763

Learning Lyapunov Functions for Piecewise Affine Systems with Neural Network Controllers

@article{Chen2020LearningLF,
  title={Learning Lyapunov Functions for Piecewise Affine Systems with Neural Network Controllers},
  author={Shaoru Chen and Mahyar Fazlyab and Manfred Morari and George J. Pappas and Victor M. Preciado},
  journal={arXiv: Optimization and Control},
  year={2020}
}
We propose an iterative method for Lyapunov-based stability analysis of piecewise affine dynamical systems in feedback with piecewise affine neural network controllers. In each iteration, a learner uses a collection of samples of the closed-loop system to propose a Lyapunov function candidate by solving a convex program. The learner then queries a verifier, which then solves a mixed-integer program to either validate the proposed Lyapunov function candidate or reject it with a counterexample, i… 

Figures from this paper

Stability analysis of complementarity systems with neural network controllers

This paper introduces a method to represent neural networks with rectified linear unit (ReLU) activations as the solution to a linear complementarity problem and turns the stability verification problem into a linear matrix inequality (LMI) feasibility problem.

Learning Stability Certificates from Data

It is demonstrated empirically that certificates for complex dynamics can be efficiently learned, and that the learned certificates can be used for downstream tasks such as adaptive control.

A Piecewise Learning Framework for Control of Unknown Nonlinear Systems with Stability Guarantees

A piecewise learning framework for controlling nonlinear systems with unknown dynamics is proposed where each piece is responsible for locally learning and controlling over some region of the domain and the runtime results are provided to demonstrate potentials of the proposed framework in real-world implementations.

Learning stability of partially observed switched linear systems

This paper proposes an algorithm for deciding stability of switched linear systems under arbitrary switching based purely on observed output data that essentially relies on an output-based Lyapunov stability framework and returns an estimate of the joint spectral radius (JSR).

Approximate Dynamic Programming for Constrained Linear Systems: A Piecewise Quadratic Approximation Approach

This paper introduces an approach combining the two methodologies to overcome their individual limitations, and proposes an ADP method for CLQR problems using Model predictive control and a novel convex and piecewise quadratic neural network.

Sparse Polynomial Optimisation for Neural Network Verification

Sparse polynomial optimisation theory and the Positivstellensatz are used, which exploits the natural cascading structure of the neural network using ideas from chordal sparsity while the later asserts the emptiness of a semi-algebraic set with a nested family of tests of non-decreasing accuracy to provide tight bounds.

Joint Differentiable Optimization and Verification for Certified Reinforcement Learning

This work proposes a framework that jointly conducts reinforcement learning and formal verification by formulating and solving a novel bilevel optimization problem, which is differentiable by the gradients from the value function and certificates.

Revisiting PGD Attacks for Stability Analysis of Large-Scale Nonlinear Systems and Perception-Based Control

This paper tailor the projected gradient descent (PGD) attack method developed in the adversarial learning community as a general-purpose ROA analysis tool for large-scale nonlinear systems and end-to-end perception-based control and proposes a finite-difference PGD estimate.

Adversarially Robust Stability Certificates can be Sample-Efficient

It is shown that under suitable assumptions of incremental stability on the underlying system, the statistical cost of learning an adversarial stability certificate is equivalent, up to constant factors, to that of learning a nominal stability certificate.

Robust Stability of Neural-Network Controlled Nonlinear Systems with Parametric Variability

A theory for stability and stabilizability of a class of neural-network controlled nonlinear systems, where the equilibria can drift when parametric changes occur is developed and a Lyapunov based convex stability certificate is developed.

References

SHOWING 1-10 OF 86 REFERENCES

A Lyapunov method for stability analysis of piecewise-affine systems over non-invariant domains

An algorithm based on linear programming is proposed, in order to prove exponential stability of the origin and to find a positively invariant estimate of its region of attraction, which is in general discontinuous on the boundaries of the regions.

Learning control lyapunov functions from counterexamples and demonstrations

This approach is able to synthesize relatively simple polynomial control Lyapunov-like functions, and in that process replace the MPC using a guaranteed and computationally less expensive controller.

Stability Analysis Using Quadratic Constraints for Systems With Neural Network Controllers

Two stability theorems are given to prove asymptotic stability and to compute an ellipsoidal inner-approximation to the region of attraction and both results rely on semidefinite programming to compute Lyapunov functions and inner-estimates of the area of attraction.

Simulation-guided lyapunov analysis for hybrid dynamical systems

This work presents a technique for discovering Lyapunov functions and barrier certificates for nonlinear and hybrid dynamical systems using a search-based approach that can be applied to a broad class of nonlinear Dynamical systems, including hybrid systems and systems with polynomial and even transcendental dynamics.

Formal Synthesis of Lyapunov Neural Networks

This work proposes a method that finds Lyapunov functions fully automatically—using machine learning—while also providing formal guarantees—using satisfiability modulo theories (SMT), and synthesises Lyap unov functions faster and over wider spatial domains than the alternatives, yet providing stronger or equal guarantees.

A survey on stability analysis of discrete-time piecewise affine systems

This survey introduces the most promising existing methods and adapts them to discrete-time PWA systems and investigates the computation of common quadratic, common quartic, piecewise affine, pieceswise quadratics and piecewise quartic Lyapunov functions using linear programming, semi-definite programming and sum-of-squares techniques.

Neural network-based control design: an LMI approach

  • S. LimanondJ. Si
  • Mathematics
    Proceedings of the 1998 American Control Conference. ACC (IEEE Cat. No.98CH36207)
  • 1998
A neural-network-based control design for a discrete-time nonlinear system with a multilayer perceptron of which the activation functions are of the sigmoid type symmetric to the origin and the stability of the closed-loop is guaranteed.

Automated and Sound Synthesis of Lyapunov Functions with SMT Solvers

The overall technique is evaluated over a broad set of benchmarks, which shows that this methodology not only scales to 10-dimensional models within reasonable computational time, but also offers a novel soundness proof for the generated Lyapunov functions and their domains of validity.

An approach to stability criteria of neural-network control systems

  • Kazuo Tanaka
  • Engineering, Mathematics
    IEEE Trans. Neural Networks
  • 1996
A method of parameter region (PR) representation, which graphically shows the location of parameters of nonlinear systems, is proposed by introducing new concepts of vertex point and minimum representation and an important theorem, useful for effectively finding a Lyapunov function, is derived.

Approximating Explicit Model Predictive Control Using Constrained Neural Networks

A modified reinforcement learning policy gradient algorithm is introduced that utilizes knowledge of the system model to efficiently train the neural network and guarantees that the network generates feasible control inputs by projecting onto polytope regions derived from the maximal control invariant set of theSystem.
...