In this demonstration, I want to go through some relevant aspects of GM’s new vehicle architecture, and how we extended the COVESA vsomeip project to support the ADAS use cases in that architecture. The advanced ADAS use cases include data collection for model training as well as eyes off autonomous driving. In that context, we use SOME/IP to talk to our peripheral ECUs, and an internally authored, open-source middleware named subspace to do communication within our ADAS SoCs. The COVESA vsomeip library, largely authored by BMW, provided a great starting point for us. However, for our ADAS needs it fell short in key areas: Its implementation was a lot more CPU intensive than we desired. There were many copies, while some were avoidable, others were not because of the design choices. Removing these copies became extremely important when payloads included large data such as Radar and Lidar frames. Its design involving proxies and a central routing manager posed safety concerns by becoming a central point of failure It lacked a serialization library that went beyond serialization and deserialization of headers I plan on presenting the modified architecture and demonstrating the new functionality as well as the performance gains. All the code I plan to present shall be made open source by the time of the conference.