• Corpus ID: 245502582

Learning Linear Complementarity Systems

  title={Learning Linear Complementarity Systems},
  author={Wanxin Jin and Alp Aydinoglu and Mathew Halm and Michael Posa},
This paper investigates the learning, or system identification, of a class of piecewiseaffine dynamical systems known as linear complementarity systems (LCSs). We propose a violation-based loss which enables efficient learning of the LCS parameterization, without prior knowledge of the hybrid mode boundaries, using gradient-based methods. The proposed violation-based loss incorporates both dynamics prediction loss and a novel complementarity violation loss. We show several properties attained… 

Figures from this paper


Identification of piecewise affine systems via mixed-integer programming
Adaptive Piecewise–Affine Inverse Modeling of Hybrid Dynamical Systems
A novel PWA identification algorithm based on convex optimization applicable to both state–space and input/output models and a novel control framework that adaptively inverts the dynamics of such systems using noisy sampled data is proposed.
On the dynamic analysis of piecewise-linear networks
A mathematically precise framework is developed that formalizes the mixed discrete and continuous behavior of these switched networks and the fundamental question of well-posedness is studied and additional properties of the behavior are derived.
Control of systems integrating logic, dynamics, and constraints
Sensitivity analysis for nonlinear programming using penalty methods
A theoretical basis is established for utilizing a penalty-function method to estimate sensitivity information of a localsolution and its associated Lagrange multipliers of a large class of nonlinear programming problems with respect to a general parametric variation in the problem functions.
Equivalence of hybrid dynamical models
A Semi-Supervised Learning Approach for Identification of Piecewise Affine Systems
This paper develops a modified self-training support vector machine algorithm to simultaneously identify the hyperplanes and parameter of each submodel with the ouputs of the cluster-based algorithm.
Piecewise linear optimal controllers for hybrid systems
A stabilizing controller is obtained by designing a model predictive controller, which is based on the minimization of a weighted l/sub 1///spl infin/-norm of the tracking error and the input trajectories over a finite horizon.