NeuReach: Learning Reachability Functions from Simulations

Abstract We present , a tool that uses neural networks for predicting reachable sets from executions of dynamical system. Unlike existing reachability tools, computes function outputs an accurate over-approximation the set any initial in parameterized family. Such functions are useful online monitoring, verification, and safe planning. implements empirical risk minimization learning functions. discuss design rationale behind optimization problem establish computed output is probably approximately correct. Our experimental evaluations over variety systems show promise. can learn complex nonlinear systems, including some beyond methods. From learned function, arbitrary reachtubes be milliseconds. available at https://github.com/sundw2014/NeuReach .