In this Perspective, we examine changes being considered to the Federal Motor Vehicle Safety Standard (FMVSS) exemption process that are aimed at facilitating deployment of autonomous vehicles (AVs). We find that, although FMVSS pose an obstacle to the deployment of many AV designs, the proposed solution — an adaption of the existing exemption process — is not well suited to AVs because traditional methods of evaluating vehicle safety as a precondition to granting exemptions are largely inapplicable. Additionally, the proposed approach may increase the public's exposure to risk by greatly raising the limit on the number of exempted vehicles using this new technology, and may add to the burdens on regulators by raising the number of exemption requests.
We instead suggest an approach in which AVs are introduced gradually as the vehicles meet a set of incremental, performance-based benchmarks. We then propose a paradigm to resolve the two questions central to this graduated approach: What should the performance benchmarks be and how many vehicles should be allowed at each benchmark? We show the mathematical relationship between these two questions and suggest that the target benchmark of AV performance can determine the cap on vehicles' deployment or, conversely, the number of vehicles desired can determine what benchmark should be set. Such an approach balances the need for measuring AV safety performance with risk exposure to the public, while facilitating the deployment of these important new technologies.