I am concerned that other than extremely geofenced situations autonomous vehicles will not exist or may not for a very long time because the base technology of the sensor systems is not sound. Can someone show me that Camera or LiDAR technology will EVER have high reliability in bad weather and/or varied surface and lighting conditions? Show me each of them spotting a fallen pole or even power line in a blizzard with fog in night and day. Then with the sun on the horizon directly in front. Or flooding conditions especially where the system would have to discern depth. Show me that the technology can ever deal with the worst case scenarios.
Making the systems see extremely well in favorable conditions could mean nothing almost nothing if you wind up so geofenced the use cases are severely restricted and you can’t sell enough product to be profitable. I believe this situation, and the untenable process of using public shadow driving for AI vs aerospace level simulation, is a casualty of Agile software development process. Where developers start from simple use cases and work toward the hard. That could be fine in simple systems where the base technology is well founded. It could be a nightmare in complex systems and where the base technology is not proven.
There is a reason DoD doesn’t use LiDAR as a primary sensor. I believe the answer here may be 3D radar. But even that may be too costly to scale. And even if you get that right what would you use to verify the radar’s data? Another 3D radar? Coupled with handover/L2/L3 being dangerous and not being able to be reliably mitigated by any monitoring and notification system you must skip these activities/levels. Especially any activity that rests control of steering for any amount of time and go straight from augmented driving to full autonomous driving. Imagine a plausible worst case scenario. The vehicle is cruising along with the AV steering. The vehicle comes up on a fallen pole, power line, flooded area or storm. The vehicle may never even try you handover to you. Or by the time it does you do not have time to gain the proper situational awareness to deal with it.
All of this is yet another reason why governments, insurers, oversight groups, the press, public etc have to demand a minimum testable scenario matrix is created. We have no idea what capabilities these vehicles actually have. With so many of these developers getting their training wheels in this type of engineering and companies embellishing their capabilities to appear to be ahead of the competition we simply cannot trust the right things are being done.
For more detail on the issues with handover, L2/L3 and public shadow driving for AI ve aerospace level simulation please see my article
Autonomous Levels 4 and 5 will never be reached without Simulation vs Public Shadow Driving for AI