A Literature Review of Simulation Fidelity for Autonomous-Vehicle Research and Development

Abstract

<div>This article explores the value of simulation for autonomous-vehicle research and development. There is ample research that details the effectiveness of simulation for training humans to fly and drive. Unfortunately, the same is not true for simulations used to train and test artificial intelligence (AI) that enables autonomous vehicles to fly and drive without humans. Research has shown that simulation “fidelity” is the most influential factor affecting training yield, but <i>psychological fidelity</i> is a widely accepted definition that does not apply to AI because it describes how well simulations engage various cognitive functions of human operators. Therefore, this investigation reviewed the literature that was published between January 2010 and May 2022 on the topic of simulation <i>fidelity</i> to understand how researchers are defining and measuring simulation <i>fidelity</i> as applied to training AI. The results reported herein illustrate that researchers are generally using agreed-upon terms such as <i>physical fidelity</i>, but there is an emerging definition of <i>functional fidelity</i> that is being adopted to replace the concept of <i>psychological fidelity</i> for training AI instead of humans.</div>