Brad, please understand I was not trying to constrain the future state to the current state. I’m just trying to express that the models being used today for autonomous development cannot deliver a level 4 autonomous vehicle because the implementation models lack the capacity to deliver the network services required for that level 4 autonomy.

This is a description of the fundamental limitation of Google’s pod-car dependency on an extremely high-speed, low latency connection to a database of high quality machine-readable interpretations of the built world for comparison to the real-time machine model generated by the sensor fusion on the vehicle. Their highway implementations don’t need this because the environment is much more predictable and controlled, but the complex urban navigation necessary for first mile / last mile connectivity demand this level of detail in Google’s current model.

That model requires a high-speed, low-latency connection to a high quality machine readable database that simply doesn’t exist for 99.9999% of the world. The leap is in how Google either eliminates that dependency, or how they support that dependency across a large geophysical area. Both are super tough!

Certainly Google’s model can deliver a level 3 implementation, but the gap between level 3 and level 4 is vast. The difference between level 3 and level 4 is like the difference between a cordless phone and a cell phone.

Either Google will have to change their model to remove those dependencies, or they will have to support these dependencies. Since neither removing the dependencies nor supporting the dependencies are part of their current deployment plan (to my knowledge), then the logical conclusion is that their deployment plan cannot provide a level 4 service.

Until Google and other autonomous car developers are willing to acknowledge their need for a network, level 4 autonomy provided by the models currently being used by companies developing autonomous vehicles will always be isolated edge-cases and not a ubiquitous or even commonplace service level.

Please see my response to Robin here for further explanation. Unfortunately I went and got involved in this discussion before my article explaining this issue in detail is ready for print so I’d like to refer everyone to a longer form explanation but I don’t have it ready for print yet.