Slashdot Mirror
Selling Full Autonomy Before It's Ready Could Backfire For Tesla (arstechnica.com)
An anonymous reader quotes a report from Ars Technica: Tesla has an Autopilot problem, and it goes far beyond the fallout from last month's deadly crash in Mountain View, California. Tesla charges $5,000 for Autopilot's lane-keeping and advanced cruise control features. On top of that, customers can pay $3,000 for what Tesla describes as "Full Self-Driving Capability." "All you will need to do is get in and tell your car where to go," Tesla's ordering page says. "Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs and roundabouts, and handle densely packed freeways with cars moving at high speed." None of these "full self-driving" capabilities are available yet. "Self-Driving functionality is dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction," the page says. "It is not possible to know exactly when each element of the functionality described above will be available, as this is highly dependent on local regulatory approval."
But the big reason full self-driving isn't available yet has nothing to do with "regulatory approval." The problem is that Tesla hasn't created the technology yet. Indeed, the company could be years away from completing work on it, and some experts doubt it will ever be possible to achieve full self-driving capabilities with the hardware installed on today's Tesla vehicles. "It's a vastly more difficult problem than most people realize," said Sam Abuelsamid, an analyst at Navigant Research and a former auto industry engineer. Tesla has a history of pre-selling products based on optimistic delivery schedules. This approach has served the company pretty well in the past, as customers ultimately loved their cars once they ultimately showed up. But that strategy could backfire hugely when it comes to Autopilot.
But the big reason full self-driving isn't available yet has nothing to do with "regulatory approval." The problem is that Tesla hasn't created the technology yet. Indeed, the company could be years away from completing work on it, and some experts doubt it will ever be possible to achieve full self-driving capabilities with the hardware installed on today's Tesla vehicles. "It's a vastly more difficult problem than most people realize," said Sam Abuelsamid, an analyst at Navigant Research and a former auto industry engineer. Tesla has a history of pre-selling products based on optimistic delivery schedules. This approach has served the company pretty well in the past, as customers ultimately loved their cars once they ultimately showed up. But that strategy could backfire hugely when it comes to Autopilot.
I also thought about how hard it would be to make a real autonomous vehicle that worked under all conditions
Getting to 90% has been done. Good weather, good visibility, few unexpected hazards
Getting to 95% will be harder, and it gets exponentially harder as you asymptotically approach 100%
The billion dollar question is...How close is close enough?
No matter how good it gets, someone will always sue, claiming it isn't perfect
The law needs to be adjusted to accept the reality that nothing is perfect
The people who think it takes hard AI to achieve full autonomy in a self driving car vastly overestimate the cognitive abilities of human drivers. The computer does not need a complete and entirely correct model of the environment to be a better driver than a person. People are very easily overwhelmed by complex traffic situations and make tons of mistakes. Roads are designed to enable safe traffic regardless of these cognitive deficiencies. Computers can take advantage of that too.
I own a Tesla and the operating costs are substantially less than any fossil fueled car I have ever owned. Charging is much less expensive than gas and the only real maintenance is replacing the tires. I normally don't charge the battery past 200 miles unless I'm going on a long trip and my typical home charging time is less than an hour. The customer service absolutely blows away any other car brand I've owned and the car just keeps getting better with new software updates. I have never heard of any of the major auto manufacturers doing updates at all for anything that wasn't a serious problem and certainly not free over-the-air updates. I paid around $70K after the tax rebate which is not even close to the high-end for a luxury car.
The people who think it takes hard AI to achieve full autonomy in a self driving car vastly overestimate the cognitive abilities of human drivers. The computer does not need a complete and entirely correct model of the environment to be a better driver than a person. People are very easily overwhelmed by complex traffic situations and make tons of mistakes. Roads are designed to enable safe traffic regardless of these cognitive deficiencies. Computers can take advantage of that too.
And people who make claims such as yours tend to forget that roads were designed for humans, not computers, and that some things that humans do very easily are very difficult for computers to do (and of course, vice-versa).
If we had roads which were designed for computers, I am sure computers would very quickly outperform human drivers. The problem is - we don't.
I believe that self-driving efforts are focused on the wrong thing: trying to reproduce a human driver, and claiming it's better if it, on average, messes up less than a human. Instead, it should be focused on trying to create an infrastructure that supports self-driving vehicles. Does that mean they will be able to drive on every imaginable road? Probably not...but likely on 95% of roads (such e.g. all roads and streets in cities), once the roll-out is complete. In this case, it probably will be safer...but if it satisfies a bunch of conditions first.
The other problem that people who claim "humans are not that smart, computers are better than them on average" miss is the variability among humans. No human driver is identical; almost all make mistakes, but the types of mistakes that are made (and the situations which they are made in) can differ widely. However, software flaws are replicated identically across many units...potentially millions of them. Joe in Chicago being a bad driver does not affect Bob in Cleveland. An undetected bug in Tesla's Autopilot will affect all Tesla owners in the world potentially, and probably under the same (or very similar) conditions. You might be in a situation where you have a 100% probability of a screw-up under certain conditions. That is simply not the case with any human driver. Now imagine if say, 20% of the cars on the road are self-driving...what you are potentially setting things up for is a black swan type of event: software-driven cars may be much safer 99% of the time, but could be prone to major screw-ups 1% of the time that will dwarf the combined effects of bad human driving.
Variability among human drivers also allows for evolutionary selection: reckless drivers will typically die, or have their licenses taken away from them. This does not remove all of the bad drivers - but does remove a great deal of them over the long run. How do we do that with self-driving cars? A destruction of one self-driving car with a flaw will not end it, because then likely all other cars of the same model and series have the same flaw. Removing just that particular car won't do it, you'd have to do a recall...now recalls can already get quite bad, imagine all of the recalls we're gonna have with self-driving vehicles (where the authorities are bound to more paranoid)...and it won't all be software problems you can patch, there will be hardware problems too.
I'm not saying these are all insurmountable problems. They can be addressed. I'm just saying there's a lot more to autonomous vehicles than the techno-optimistic "self-driving cars just need to cause less accidents on average than humans" stance. A lot more. It cannot be reduced just to a single metric. Or just a bunch of numeric metrics.
I happen to have a Tesla S and I do love the car.
They are a joy to drive. The acceleration is second to none. The handling is great, almost on par with the ( now deprecated ) hydraulic steering on BMWs's I used to drive. The cabin technology is great. I also really like the fact that the car gets free updates that actually improve it on a continuous basis!
I also feel extra pleased with my purchase because I get free EV charging in my office building, premium EV parking at most malls, and I just learned that Tesla will install a supercharger ( which I also get for free ) at my local mall. Heck I even get premium parking and free power at Ikea! ( I would get these with a lower cost EV but their really just a 'feel good' bonus. )
The autpilot is def misnamed. It's nowhere near autonomous. Just now it's more like a junior co-pilot. Would I like it to be better? Yes I sure would. But I will say that once you know it's limitations it is very helpful and - contrary to popular opinion - it does an excellent job in all weather conditions except heavy snow which covers the road. I'd say 70% of my daily commute is now handled by autopilot.