Reliably-stabilizing piecewise-affine neural network controllers

A common problem affecting neural network (NN) approximations of model predictive control (MPC) policies is the lack of analytical tools to assess the stability of the closed-loop system under the action of the NN-based controller. We present a general procedure to quantify the performance of such a controller, or to design minimum complexity NNs with rectified linear units (ReLUs) that preserve the desirable properties of a given MPC scheme. By quantifying the approximation error between NN-based and MPC-based state-to-input mappings, we first establish suitable conditions involving two key quantities, the worst-case error and the Lipschitz constant, guaranteeing the stability of the closed-loop system. We then develop an offline, mixed-integer optimization-based method to compute those quantities exactly. Together these techniques provide conditions sufficient to certify the stability and performance of a ReLU-based approximation of an MPC control law.

[1]  Olaf Stursberg,et al.  Polytopic Input Constraints in Learning-Based Optimal Control Using Neural Networks , 2021, 2021 European Control Conference (ECC).

[2]  E. Yaz.  Linear Matrix Inequalities In System And Control Theory , 1998, Proceedings of the IEEE.

[3]  Razvan Pascanu,et al.  On the Number of Linear Regions of Deep Neural Networks , 2014, NIPS.

[4]  E. T. Maddalena,et al.  A Neural Network Architecture to Learn Explicit MPC Controllers from Data , 2019, ArXiv.

[5]  Alexandros G. Dimakis,et al.  Exactly Computing the Local Lipschitz Constant of ReLU Networks , 2020, NeurIPS.

[6]  Indranil Saha,et al.  DeepControl: Energy-Efficient Control of a Quadrotor using a Deep Neural Network , 2019, 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

[7]  Alberto Bemporad,et al.  The explicit linear quadratic regulator for constrained systems , 2003, Autom..

[8]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[9]  Mark Cannon,et al.  Some observations on the activity of terminal constraints in linear MPC , 2016, 2016 European Control Conference (ECC).

[10]  Sergey Levine,et al.  Learning deep control policies for autonomous aerial vehicles with MPC-guided policy search , 2015, 2016 IEEE International Conference on Robotics and Automation (ICRA).

[11]  Monimoy Bujarbaruah,et al.  Near-Optimal Rapid MPC Using Neural Networks: A Primal-Dual Policy Learning Framework , 2019, IEEE Transactions on Control Systems Technology.

[12]  Tor Arne Johansen,et al.  Approximate explicit constrained linear model predictive control via orthogonal search tree , 2003, IEEE Trans. Autom. Control..

[13]  Pushmeet Kohli,et al.  A Unified View of Piecewise Linear Neural Network Verification , 2017, NeurIPS.

[14]  Charles Richter,et al.  Bayesian Learning for Safe High-Speed Navigation in Unknown Environments , 2015, ISRR.

[15]  Alberto Bemporad,et al.  Control of systems integrating logic, dynamics, and constraints , 1999, Autom..

[16]  A. Bemporad,et al.  Suboptimal Explicit Receding Horizon Control via Approximate Multiparametric Quadratic Programming , 2003 .

[17]  V. L. Wisniewski,et al.  Embedded PWM Predictive Control of DC-DC Power Converters Via Piecewise-Affine Neural Networks , 2021, IEEE Open Journal of the Industrial Electronics Society.

[18]  Thomas Parisini,et al.  A receding-horizon regulator for nonlinear systems and a neural approximation , 1995, Autom..

[19]  Benjamin Karg,et al.  Efficient Representation and Approximation of Model Predictive Control Laws via Deep Learning , 2018, IEEE Transactions on Cybernetics.

[20]  K. T. Tan,et al.  Linear systems with state and control constraints: the theory and application of maximal output admissible sets , 1991 .

[21]  Jan M. Maciejowski,et al.  Predictive control : with constraints , 2002 .

[22]  S. Joe Qin,et al.  A survey of industrial model predictive control technology , 2003 .

[23]  Martin Mönnigmann,et al.  On the maximal controller gain in linear MPC , 2017 .

[24]  Franco Blanchini,et al.  Set-theoretic methods in control , 2007 .

[25]  Manfred Morari,et al.  Safety Verification and Robustness Analysis of Neural Networks via Quadratic Constraints and Semidefinite Programming , 2019, ArXiv.

[26]  David Q. Mayne,et al.  Constrained model predictive control: Stability and optimality , 2000, Autom..

[27]  Joel A. Paulson,et al.  Approximate Closed-Loop Robust Model Predictive Control With Guaranteed Stability and Constraint Satisfaction , 2020, IEEE Control Systems Letters.

[28]  Matteo Fischetti,et al.  Deep neural networks and mixed integer linear optimization , 2018, Constraints.

[29]  T. Johansen,et al.  On explicit suboptimal LQR with state and input constraints , 2000, Proceedings of the 39th IEEE Conference on Decision and Control (Cat. No.00CH37187).

[30]  Benjamin Karg,et al.  Stability and feasibility of neural network-based controllers via output range analysis , 2020, 2020 59th IEEE Conference on Decision and Control (CDC).

[31]  Alberto Bemporad,et al.  Predictive Control for Linear and Hybrid Systems , 2017 .

[32]  Weiming Xiang,et al.  Reachable Set Estimation and Safety Verification for Piecewise Linear Systems with Neural Network Controllers , 2018, 2018 Annual American Control Conference (ACC).

[33]  G. Birkhoff,et al.  Piecewise affine functions and polyhedral sets , 1994 .

[34]  Frank Allgöwer,et al.  Safe and Fast Tracking on a Robot Manipulator: Robust MPC and Neural Network Control , 2020, IEEE Robotics and Automation Letters.

[35]  Alberto Bemporad,et al.  A survey on explicit model predictive control , 2009 .

[36]  Zhong-Ping Jiang,et al.  Input-to-state stability for discrete-time nonlinear systems , 1999 .

[37]  Venkatesh Saligrama,et al.  Piecewise Linear Regression via a Difference of Convex Functions , 2020, ICML.

[38]  Tor Arne Johansen,et al.  Toward Dependable Embedded Model Predictive Control , 2017, IEEE Systems Journal.

[39]  Vijay Kumar,et al.  Approximating Explicit Model Predictive Control Using Constrained Neural Networks , 2018, 2018 Annual American Control Conference (ACC).

[40]  R. Agarwal,et al.  Difference Equations and Inequalities: Theory, Methods, and Applications , 2019 .

[41]  Miroslav Fikar,et al.  Clipping-Based Complexity Reduction in Explicit MPC , 2012, IEEE Transactions on Automatic Control.

[42]  Frank Allgöwer,et al.  Learning an Approximate Model Predictive Controller With Guarantees , 2018, IEEE Control Systems Letters.

[43]  Raman Arora,et al.  Understanding Deep Neural Networks with Rectified Linear Units , 2016, Electron. Colloquium Comput. Complex..

[44]  David Rolnick,et al.  Deep ReLU Networks Have Surprisingly Few Activation Patterns , 2019, NeurIPS.

[45]  Per-Olof Gutman,et al.  Convergence of an algorithm to find maximal state constraint sets for discrete-time linear dynamical systems with bounded controls and states , 1985 .

[46]  Meixin Zhu,et al.  Safe, Efficient, and Comfortable Velocity Control based on Reinforcement Learning for Autonomous Driving , 2019, ArXiv.

[47]  Manfred Morari,et al.  Approximate explicit MPC using bilevel optimization , 2009, 2009 European Control Conference (ECC).

[48]  Yuval Tassa,et al.  An integrated system for real-time model predictive control of humanoid robots , 2013, 2013 13th IEEE-RAS International Conference on Humanoid Robots (Humanoids).

[49]  Kurt Hornik,et al.  Approximation capabilities of multilayer feedforward networks , 1991, Neural Networks.