Flight Risk
Member
So Tesla doesn't do racing. Disappointing. It's a shame that they won't be represented at the Nikola Tesla EV Rally.
You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
I'm equally skeptical about the timing, but I completely disagree about Elon underestimating the challenge. Some of the biggest challenges at SpaceX are avionics/guidance software. He is surrounded by some of the best controls engineers in the world at SpaceX, and he's likely gotten his hands dirty with some of the design details in that area too.
I suspect the intent is to not to address the exceptions, but just to address the simple case. The human will be required to stay alert and take over when an exception occurs just like human override on cruise control.It isn't the boring, follow the guy in front you driving that is hard, it is all the exceptions.
Yes, but the environment that a rocket works in is FAR less complex that a car on the freeway. Radar isn't high enough resolution to understand what is actually happening. You need vision for that. And computer vision understanding is still an unsolved AI problem.
Question, how is a car going to understand when a cop starts zip zagging across the freeway to do a full freeway hot stop?
When traffic cones are placed on the road to get everyone moved off a lane, how will the car see and understand the cones?
When an accident occurs and flares are set around a crash that just occurred, how will the car understand the scene and react accordingly?
When you are driving through a snow storm or a very heavy rain, how will the car deal with the crappy vision?
It isn't the boring, follow the guy in front you driving that is hard, it is all the exceptions.
Fair point, I guess you kind of have to believe it when you see it. But still when I hear "less than a year" that coincides nicely with the MX ramp up, so I'm kind of wondering if the X delays have been partially dependent on the autopilot progress...if he really wants to knock people's socks off that seems like a pretty good way to do it, but I don't know, maybe all luxury cars will be coming out with that next year.
When traffic cones are placed on the road to get everyone moved off a lane, how will the car see and understand the cones?
When an accident occurs and flares are set around a crash that just occurred, how will the car understand the scene and react accordingly?
Yes, but the environment that a rocket works in is FAR less complex that a car on the freeway. Radar isn't high enough resolution to understand what is actually happening. You need vision for that. And computer vision understanding is still an unsolved AI problem.
Question, how is a car going to understand when a cop starts zip zagging across the freeway to do a full freeway hot stop?
When traffic cones are placed on the road to get everyone moved off a lane, how will the car see and understand the cones?
When an accident occurs and flares are set around a crash that just occurred, how will the car understand the scene and react accordingly?
When you are driving through a snow storm or a very heavy rain, how will the car deal with the crappy vision?
It isn't the boring, follow the guy in front you driving that is hard, it is all the exceptions.
Well, at least for these two things, I think with the car knowing it's location + getting traffic updates since it's always connected to the internet, will make it easy to correct itself on the road.
Computer vision and AI are two separate parts of this. With respect to the vision itself (physical sensor hardware and low level interpretation of the inputs) it's not unsolved, simply unsolved thus far at a low enough cost - see Google's LIDAR solution. The AI side of it is definitely a bigger challenge.
But what you're saying is exactly with what Elon has said previously, that the first 90% is relatively easy, then dealing with exceptions to get to 99% is exponentially harder, 99.9% exponentially harder than that, etc. That's precisely why they are making it analogous to aircraft auto-pilot (where a human is still required regardless) rather than full autonomy. Weather conditions is the biggest single challenge for current sensor technology, followed by higher level software reliably understanding unusual road conditions like you mentioned.
But, it's quite feasible in the next two years Tesla can implement an autopilot for freeways that works like an advanced cruise control you can use in ideal conditions, so long as they can sort out who's responsible when the inevitable guy falls asleep on the hour long freeway autopilot trip and doesn't wake up to take back control when required.
Sounds like he wants to release tesla's patents....
Nah it'll be trickier than that to get it reliable. For example, imagine you are the first car to come across some traffic anomaly... it wouldn't be reported yet so the AI would have to determine it on its own - a significant challenge given the scope of what may go wrong in road conditions.
I suspect the intent is to not to address the exceptions, but just to address the simple case. The human will be required to stay alert and take over when an exception occurs just like human override on cruise control.
That's true and I can see that happening, but what's the percentage that it will be the first car at the scene of a wreck as an advanced car with this type of tech? Not very high until all cars are like this. And when are cones are the road? only during accidents or construction, so why not just code the car to understand that if it sees a cone object in the distance up ahead, to go ahead and move over to another lane when it can.
My concern is that a "freeway autopilot" capability will almost certainly result in some drivers thinking that they don't have to watch the road as carefully and they will be tempted to do other things even more than they do now: answer text or email on their phone (despite that not being legally allowed), put on makeup, and who knows what else!
I read a piece in the MIT online blog saying this was the exact concern that Google has - their initial plan was a 'freeway pilot' but they quickly found that the 'driver' was so inattentive that the option of bailing out and handing back control was not really viable - hence the shift to the completely automatic, no steering wheel car they recently showed.
That is a much harder problem, but avoids the simpler case with the catastrophic end-game!
I read a piece in the MIT online blog saying this was the exact concern that Google has - their initial plan was a 'freeway pilot' but they quickly found that the 'driver' was so inattentive that the option of bailing out and handing back control was not really viable - hence the shift to the completely automatic, no steering wheel car they recently showed.
That is a much harder problem, but avoids the simpler case with the catastrophic end-game!
He didn't say "in all circumstances". It'll be the boring, follow-the-guy-in-front that comes first, along with warnings when an exception occurs and then gradually adding exception handlers.
hence google's steering-wheel-less prototype has a 25mile speed limit - to limit the catastrophic end-game.
lazy humans shaped Google's new autonomous car:
http://www.technologyreview.com/news/527756/lazy-humans-shaped-googles-new-autonomous-car/