TBH, Tesla should embrace CAV and V2I. Medium/long term its a solution to so many problems. Not sure short term if its rolling out. I use to work in a group that wrote the majority of UK traffic light low/kernel level control systems and nearly 3 years ago, the V2I functionality needed for traffic light control was well progressed. But we were not the people who packaged it up in the hardware and deployed.
V2I allows many features that is impossible to achieve with visual detection, such as having a window when the lights are guaranteed to remain at green - for many lights its not a fixed window and can change from phase to phase so cannot be predicted. Lots of other benefits with V2I and being on board with it provides simple solutions to what would otherwise be complex visual problems. I'm not suggesting a single solution, plenty of benefit in having visual solution too. One solution does not fit all problems. I really think Tesla should get on board with this if not already - but I never heard their name come up in discussions on the subject and I sat opposite someone on the global committee for this.