mlsanityData to 14 Oct 2026

Provably Correct Training of Neural Network Controllers Using Reachability Analysis

arXiv 2102.10806 ↗First posted 22 Feb 2021eess.SY cs.LG cs.ROcited by 5

Not on the map. Why?

Abstract (arXiv)

In this paper, we consider the problem of training neural network (NN) controllers for nonlinear dynamical systems that are guaranteed to satisfy safety and liveness (e.g., reach-avoid) properties. Our approach is to combine model-based design methodologies for dynamical systems with data-driven approaches to achieve this target. We confine our attention to NNs with Rectifier Linear Unit (ReLU) nonlinearity which are known to represent Continuous Piece-Wise Affine (CPWA) functions. Given a mathematical model of the dynamical system, we compute a finite-state abstract model that captures the closed-loop behavior under all possible CPWA controllers. Using this finite-state abstract model, our framework identifies a family of CPWA functions guaranteed to satisfy the safety requirements. We augment the learning algorithm with a NN weight projection operator during training that enforces the resulting NN to represent a CPWA function from the provably safe family of CPWA functions. Moreover, the proposed framework uses the finite-state abstract model to identify candidate CPWA functions that may satisfy the liveness properties. Using such candidate CPWA functions, the proposed framework biases the NN training to achieve the liveness specification. We show the efficacy of the proposed framework both in simulation and on an actual robotic vehicle.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 35 of its 52 references that are in the map. First posted on arXiv 22 Feb 2021.

Builds on

35 of 52 references.

References · 35

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2017Constrained Policy Optimizationcited by 983
  3. 2017Safe Model-based Reinforcement Learning with Stability Guaranteescited by 366
  4. 2018A Lyapunov-based Approach to Safe Reinforcement Learningcited by 297
  5. 2019Efficient and Accurate Estimation of Lipschitz Constants for Deep Neural Networkscited by 293
  6. 2019End-to-End Safe Reinforcement Learning through Barrier Functions for Safety-Critical Continuous Control Taskscited by 289
  7. 2019Algorithms for Verifying Deep Neural Networkscited by 221
  8. 2017A General Safety Framework for Learning-Based Control in Uncertain Robotic Systemscited by 163
  9. 2019Lyapunov-based Safe Policy Optimization for Continuous Controlcited by 145
  10. 2016Bayesian Optimization with Safety Constraints: Safe and Automatic Parameter Tuning in Roboticscited by 134
  11. 2018Learning-based Model Predictive Control for Safe Explorationcited by 120
  12. 2018A predictive safety filter for learning-based control of constrained nonlinear dynamical systemscited by 118
23 more
  1. 2017Trial without Error: Towards Safe Reinforcement Learning via Human Interventioncited by 103
  2. 2020Training robust neural networks using Lipschitz boundscited by 99
  3. 2013On the number of response regions of deep feed forward networks with piece-wise linear activationscited by 94
  4. 2020Learning Control Barrier Functions from Expert Demonstrationscited by 92
  5. 2018Neural Lander: Stable Drone Landing Control using Learned Dynamicscited by 85
  6. 2018Verisig: verifying safety properties of hybrid systems with neural network controllerscited by 84
  7. 2016Safe Exploration in Finite Markov Decision Processes with Gaussian Processescited by 75
  8. 2018Formal Verification of Neural Network Controlled Autonomous Systemscited by 41
  9. 2018Linear model predictive safety certification for learning-based controlcited by 37
  10. 2017Safe Learning of Quadrotor Dynamics Using Barrier Certificatescited by 35
  11. 2020Safe Reinforcement Learning for Autonomous Vehicles through Parallel Constrained Policy Optimizationcited by 35
  12. 2020Synthesis of Control Barrier Functions Using a Supervised Machine Learning Approachcited by 35
  13. 2019Robust Regression for Safe Exploration in Controlcited by 24
  14. 2020Learning Lyapunov Functions for Hybrid Systemscited by 24
  15. 2019Temporal Logic Guided Safe Reinforcement Learning Using Control Barrier Functionscited by 21
  16. 2020Safe Multi-Agent Interaction through Robust Control Barrier Functions with Learned Uncertaintiescited by 20
  17. 2020ShieldNN: A Provably Safe NN Filter for Unsafe NN Controllerscited by 17
  18. 2019TIRA: Toolbox for Interval Reachability Analysiscited by 12
  19. 2018Reachable Set Estimation and Verification for Neural Network Models of Nonlinear Dynamic Systemscited by 7
  20. 2020A Control Barrier Perspective on Episodic Learning via Projection-to-State Safetycited by 7
  21. 2020Two-Level Lattice Neural Network Architectures for Control of Nonlinear Systemscited by 4
  22. 2017Scalable synthesis of safety certificates from data with application to learning-based controlcited by 2
  23. 2020Adaptive Control Barrier Functions for Safety-Critical Systemscited by 1

Built on by

Cited by year

Citing papers per year: 2021: 2; 2022: 2; 2023: 0; 2024: 1.

5 papers.

  1. 2022NNLander-VeriF: A Neural Network Formal Verification Framework for Vision-Based Autonomous Aircraft Landingcited by 19
  2. 2022BERN-NN: Tight Bound Propagation For Neural Networks Using Bernstein Polynomial Interval Arithmeticcited by 5
  3. 2021Provably Safe Model-Based Meta Reinforcement Learning: An Abstraction-Based Approachcited by 1
  4. 2024Counterexample-Guided Repair of Reinforcement Learning Systems Using Safety Criticscited by 1
  5. 2021Failing with Grace: Learning Neural Network Controllers that are Boundedly Unsafecited by 0