An Old Dog with New Tricks: Lessons from Building an AV Testbed with Autoware

While building a testbed with a working Autonomous Vehicle (AV) running an open-source Autonomous Driving Stack (ADS) is essential for supporting AV-related research and experiments, it is also challenging, especially for academic researchers. In particular, academic researchers often have both limited funds to acquire the latest hardware and software components, and limited human resources and experience to integrate these components into a testbed. In this paper, we report the steps taken and lessons learned in upgrading the ADS on our 2017 Lincoln MKZ to the latest Autoware version (Universe), as well as associated efforts in localization, generating HD maps, integrating with simulators, and building digital twins. We describe the challenges encountered, including current deficiencies in Autoware Universe, and some of the solutions to overcome these challenges. We also present results obtained from our experiments and point out future research and experimental work. The source code of our work can be found at https://github.com/ub-cavas/ub-lincoln-docker. We believe that this paper could provide the community working on AV research useful insights into both building and using an AV experimental platform based on an open-source autonomous driving stack.