4 ms·
Like the uncanny valley, the edge cases for self-driving cars will take more work than the rest of the effort combined... by at least an order of magnitude. Th
by xenadu02 10y ago
Like the uncanny valley, the edge cases for self-driving cars will take more work than the rest of the effort combined... by at least an order of magnitude.
There is far more money in automating long-distance trucking. My bet is the first real use of self-driving "cars" will be for trucks. The human driver can sleep, eat a meal, etc but will remain on-board for safety's sake and dealing with unanticipated edge cases (e.g. the AI wakes up the driver when it runs into a traffic jam due to an accident. The AI can probably recognize cop cars, flares, etc but won't be able to navigate the scene of the accident while responding to instructions from police officers)
The next small step will be "follower" trucks. The lead truck has a human driver same as before, but one or two AI-driven trucks follow. They do their best to track the human driver's response to unexpected situations. Worst-case, they pull over and flag the human driver who pulls over and uses a VR rig to pilot them past the problem.
Meanwhile the application of "self-driving" technology to cars over the next 10 years will be to avoid accidents, not to actually be self-driving.
Will we have fully autonomous vehicles within 50 years? Almost certainly. Will we have them in the next 5? Almost certainly not.
- advantager 10y agoSelf driving haul trucks are already at work at a few open pit mine sites around the world. These are vehicles driving relatively short, prescribed routes, with well defined patterns. The few articles I've read seem to indicate they're operating just fine, although I'd be interested to learn more. If the cost savings are there, I'd bet the entire mining industry will move in that direction rather quickly. http://www.riotinto.com/ourcommitment/spotlight-18130_18328.aspx http://www.riotinto.com/ourcommitment/spotlight-18130_18328....
- stale2002 10y agoYou overestimate how good humans are at driving. I'd argue that the shit self driving cars we have right now are ALREADY statistically safer than humans. You don't have to be perfect. Just better than an average driver. And average drivers are drunk, sleep deprived, and texting. That is a very low bar to pass.
- majormajor 10y agoI think you underestimate the effects of weather, amount of poorly lined roads out there, etc.
- mattnewport 10y agoTo sell you need the car to be a better driver than the customer thinks they are which is a higher bar since most drivers think they are better than average.
- stale2002 10y agoDoubt it. Even if they were much more dangerous, there is still a huge market. Safely, is arguably pretty low on the list of things that consumers care about. The problem with self driving cars right now is purely political. But even that isn't insurmountable. All you have to do is release in a low government regulation red state, like Arizona. And then the rest of the country will follow, once everyone in Arizona starts using them and realize how awesome they are.
- kayoone 10y agoMaybe on highways in good weather conditions we are close, but the chaos of cities like Paris, Madrid, Rome with their narrow roads and millions of edge case situations will be a tough problems for self driving AIs.