Let me toss out an idea that is likely to be quite unpopular: train FSD to do police work.
Tesla sets up pilot projects with police departments and other law enforcement. Model 3 are fully equipped to be patrol cars, but FSD rides along to learn. Officers are trained on how to work with FSD system to capture information. For example, when an officer pulls over a car, FSD learn why the car was pulled over, year, make, model, color, and any other pertinent descriptors.
Suppose for example, an officer pulls someone over for suspected DUI. Data around this is captured. The NN can then learn what sort of driving behavior lead the officer to suspect DUI. If the DUI is confirmed, this too can help the NN differentiate between suspected and actual DUI driving behavior. For other examples, an officer pulls over a vehicle because it is driven recklessly, a tail light is not working, or a load is not properly secured to vehicle. Again the NN learns to recognize all of these situations.
So first point here is the police patrol work can train the NN to recognize all sorts of edge cases and road hazards. But it gets creeper. What the NN learns can eventually support police work. For example, being able to spot reckless or drunk driving, FSD Cop's Version can objectively score driver, document their behavior, and capture identifying features. In FSD patrol mode, the police officer can cede most of the driving to the vehicle focus on observing conditions and behavior. When an officer must pull over a driver, the vehicle also is there to observe as back up. In an extreme case where an officer is shot, FSD would be trained to recognize this, call for help, and capture evidence. In the case of policy pursuit, the vehicles can track the target vehicle and share information with other vehicles in the area, so that the police cars work seamlessly as a team to follow a suspect. Persons, vehicles and other objects that the police are searching for can be identified by FSD Cop's Version.
As a third point, building a strong relationship with local law enforcement can help local politicians to form a positive view of autonomous vehicles. So these pilot programs pave the way for favorable local policies. Law enforcement becomes comfortable with the new technology and understands the how that technology can make the streets much safer.
A fourth point, getting dangerous drivers out of the driver's seat is one of the ways autonomous vehicles can improve safety and avoid getting into collisions. Obviously we'd want FSD to know how to avoid drunk drivers. Helping the police get drunk drivers off the road is the next logical step.
And the final point is that there will be a market for this sort of thing: Police Robocars. Many cities have security camera that support police work. Integrating that in patrol cars would be a next logical step. So long as all this is pretty much inevitable, Tesla may as well position FSD well for this market.
Fire away!