Architecting Artificial Intelligence for Autonomous Cars: The OpenPilot Framework.

Openpilot is a vast open-source semi-automated driving system developed by comma.ai, with 200+ contributors and 750K lines of code according to the OpenHub open-source community-tracking portal. On the one hand, the documentation available gives insights on what Openpilot is capable of doing, how to install it and how people can contribute to it, while the development team posts periodically update on the company’s blog on the state of the project and implementation of new features. On the other hand, this material does little in helping newcomers to embrace the technology with near-zero knowledge over its intended programming model. What is more, not having access to closed-source autonomous-driving framworks does not warrant an appropriate architectural analysis of the framework, the decisions to be made, and the concerns typically emerging in such a decision-making process. This study addresses this gap by exploiting mixed-methods research, featuring (a) an ethnographical study of the community’s operations and (b) a reverse engineering of the OpenPilot codebase both from a structure and operations perspective. This paper is intended as an early-stage adopter study for those software designers, operators, policy-makers and other practitioners aiming at embracing the aforementioned framework and programming model for their own professional endeavours.