Ah, me. I knew I should have passed on posting in favor of a second glass of wine
OK, Here goes. I hate to play the credentials game, but I have worked on autonomous vehicles and sensor technology more than a little. Also taught the subject. It is not improving by “leaps and bounds” except in marketing material. The sensor fusion problem is as intractable as it always was. Sensor x says one thing, sensor y says something different, who do you believe? When one asks why anybody thinks a robot would be better than this than a human in all circumstances the response usually is 'blah blah artificial intelligence blah blah machine learning blah blah." No, show me your algorithm, and describe your verification and validation process (another subject I used to teach).
The statistics on accidents are both general and misleading. The great (and unquantified) majority of accidents are multifactor in nature, and we do not know what the weight of “inattentive and incompetent” drivers is in the mix. But it’s a nice justification for marketing hype aimed at people who do not understand or appreciate that human error is a symptom and not a cause.
The “obvious” conclusion that driverless cars will have a significantly impact on the accident rate is wholly unproven by anything that approaches detailed and rigorous analysis. Remember, driverless cars are being developed under the Silicon Valley business model: overpromise, rush to deployment, and let the users find the bugs. Works fine for video games and other toys, not so good when the thing you are deploying can kill people.
There are two things that are consistently missing from the marketing hype: opportunity cost and adversarial analysis. Opportunity cost says that since resources are always limited, you should not only take into consideration the positive aspects of option A but also the negative aspects of not doing options B, C, and D because you’ve spent your resources on A. Nobody has, or wants to in the Silicon Valley model, bring up the question of much the accident rate could be reduced if the resources being spent on autonomous vehicles were instead directed to other factors in the causal mix.
Adversarial analysis says you look at “how can this be abused?” Nobody wants to talk about that in the Silicon Valley business model (might make the venture capitalists nervous). Remember, Facebook and its unregulated capture and aggregation of personal data was going to be an unalloyed benefit to mankind and we now see how that turned out. Think about a supply-chain attack on the opaque, largely unverifiable software load which has full and irrevocable authority over the bundle of kinetic energy that is a moving vehicle. That is just one of any number of potential vulnerabilities that could be exploited by a malicious entity.
So my advice is to enjoy the hype while you can. Reality will intrude soon enough – in fact, it’s already poking its nose through the door.
Cheers,
Earl