Simulation generation using temporal logic

Abstract

Techniques for determining simulations to confirm programmatic logic are discussed herein. Such simulations may be used to identify errors in programmatic logic. As an example, a system may simulate an autonomous vehicle operating in an environment by setting various initialization parameters. Temporal logic, such as Linear Temporal Logic (LTL) and/or Signal Temporal Logic (STL) may be used to determine a numeric cost associated with how closely one or more policies are violated for each simulation of a group of simulations. Based on the costs computed, additional sets of simulations may be created using an evolutionary algorithm. Flaws in programmatic logic controlling the system may be identified based on the evolutionary algorithms and cost defined.