You don’t sufficiently address the potential costs. For example, if a city can use the sensors to conduct traffic research can they not also be used for surveillance?
It also occurs to me that arguing we need road sensors for driverless cars makes you in some sense an AI pessimist. After all, humans are proof of concept that all you need to drive reasonably well are two sensors called eyes connected to a sufficiently sophisticated visual system and brain for making the judgments that pass on to the motor system.
Yet I don’t think we need to get to “human level AI” to get to “human level driving AI”. We might even get to the latter pretty soon. So while external sensors would improve things in the interim they don’t at all seem to be required in the medium run.