Skip to content

Overview

The TrustworthySearch API is an efficient risk analysis service for analyzing autonomous vehicle (AV) technologies. The service acts as an optimization layer on top of simulation by choosing which scenarios to run, providing statistical coverage over failure modes, and prioritizing failures by likelihood. Our search algorithms efficiently generate useful data and actionable insights regarding safety. Most importantly, TrustworthySearch interacts with AV systems in a black-box manner, preserving the integrity and confidentiality of the software under test.

Checklist

At a high-level the basic steps in integrating TrustworthySearch are as follows:

  • Determine base scenario e.g. unprotected left at a T-junction

  • Determine whether the simulation is pre- or post-perception

  • Identify if the scenario execution is deterministic

  • Determine environment-vehicle parameterizations

  • Write scenario parameterization as a list of random variables

  • Decide on a set of laws which describe the specification

  • Identify any laws which have binary semantics

  • Determine the metric(s) that will be used to score simulations

  • Determine exit conditions and how they will affect the satisfaction of the specification

  • Determine a rough estimate of the failure probability to estimate of the computational resources that will be necessary for full evaluation

  • Integrate TrustworthySearch API with your current simulation stack by building upon the TrustworthySearch demo code.