Provably Safe Reinforcement Learning via Action Projection using Reachability Analysis and Polynomial Zonotopes

no code implementations • 19 Oct 2022 • Niklas Kochdumper, Hanna Krasowski, Xiao Wang, Stanley Bak, Matthias Althoff

While reinforcement learning produces very promising results for many applications, its main disadvantage is the lack of safety guarantees, which prevents its use in safety-critical systems.

reinforcement-learning Reinforcement Learning (RL) +1

Open- and Closed-Loop Neural Network Verification using Polynomial Zonotopes

no code implementations • 6 Jul 2022 • Niklas Kochdumper, Christian Schilling, Matthias Althoff, Stanley Bak

We present a novel approach to efficiently compute tight non-convex enclosures of the image through neural networks with ReLU, sigmoid, or hyperbolic tangent activation functions.

Formal synthesis of closed-form sampled-data controllers for nonlinear continuous-time systems under STL specifications

no code implementations • 7 Jun 2020 • Cees F. Verdier, Niklas Kochdumper, Matthias Althoff, Manuel Mazo Jr

Subsequently, the best candidate is verified using reachability analysis; if the candidate solution does not satisfy the specification, an initial condition violating the specification is extracted as a counterexample.