Apologies to those who had to watch the back and forth in the ST thread the other day, I'm usually one who advocates not responding to our trollier users... but I... just... couldn't... resist
I do think autonomy is the most important technology for the future of the automobile, I also think it's clear that Tesla has a significant lead here wrt existing products. I also believe, but can't prove, that Tesla has the best team and is making the fastest progress toward autonomy; I can't prove this, but I think the following are good reasons to believe it:
1) Their existing system is the best, despite having far fewer hardware resources than some other systems.
2) Their existing system is the best despite the fact that other companies have been working on lane-keeping systems for much longer than Tesla. For example Nissan first introduced lane keeping in 2001, Toyota in 2002, Honda in 2003.
3) Tesla has, in general, made it a higher priority than other manufacturers.
4) No other auto company has equal access to software talent due to Tesla's proximity to Silicon Valley and Elon's star power.
In addition to having the best system and making the fastest progress, Tesla's ability to issue OTA updates gives them a 1-2 year advantage over competitors as I think it's clear that the hardware is easier than the software (sure, you could have the dealership update firmware, but this is pretty kludgy and will become unwieldy when new safety related fixes are constantly required). It may actually be the case that realtime OTA updates are *required* for *any* autonomous vehicle in the future as it may be that a current database of all regulatory traffic signs/signals will be required in some countries for all autonomous vehicles.
Anyway I think understanding the tech involved will help people understand exactly how plausible full autonomy is at the present time. It's going to be a big day when the first transfer of liability from owner to manufacturer takes place. Below is an update of my first post on the topic from the ST thread, which I wrote hastily while in a meeting.
...Full autonomy is coming, I think in 5 years or less, probably less. That doesn't mean full autonomy on every road on planet earth, or on every road in the United States, it only means full autonomy on some usefully large subset of roads, probably starting with US interstates. I think the following technologies in combination, are sufficient for full autonomy.
1) GPS with SBAS (and soon LBAS) reliably gives accuracy greater than 12 inches, I see this accuracy every week, I fly aircraft that have it. This in combination with high precision maps will give the vehicle great situational awareness with regards to position on the network of streets. Vehicles will have redundant GPS systems.
Tesla actually has the ability to create their own LBAS system by using preset points on roads every 20 miles or so that have a precisely known location and that are identified by the camera system, I'd guess the camera system will have accuracy to a couple inches which means that the GPS signal can be corrected to within a couple inches.
2) Redundant binocular (+) camera systems will paint a 3D picture of the surroundings, recognizing lanes, vehicles, pedestrians, animals, street signs etc. This is the hardest part, machine object recognition has only recently surpassed that of humans.
3) Redundant radar systems will paint a 3D picture of the world that complements that of the binocular (+) camera systems. They will also see through some objects that the binocular (+) camera systems cannot and will give some density information allowing the system to distinguish a plastic bag blowing in the wind from a truck tire bouncing towards your face.
4) Ultrasonic sensors give a tertiary system for verifying objects that are near the vehicle.
5) Inertial navigation system, off the shelf GPS systems only update 10x per second you can get quicker systems but they're expensive. The inertial system will smooth GPS data and fill in for it in areas with poor GPS reception. These systems can have very high sampling rates, over 1000Hz.
6) Large database of relevant data including a complete roadmap, traffic signs/signals, obstacles, lane width, number of lanes etc.
7) Massive processing power: nVidia's PX2.0 has almost exactly 1000x the processing power of the Tegra 3 currently powering AP 1.0. They will need to have Redundant processing systems. [edit] User MP3Mike says autopilot does not use the Tegra 3, this may be the case, I believe it has more processing power than any other onboard chip but perhaps not.
It seems to me that ALL the technologies required to make a vehicle autonomous already exist. They just need to be combined with the right logic, I don't think this will be as difficult as some software problems that have already been solved.
If the above is correct, or in the vicinity of correct, it has serious implications for the value of Tesla. We've all seen the speculation about how the economics of personal transportation will change, given autonomy, from Adam Jonas and others. I suspect Tesla will get an autonomous capable product to market 3 or more years before the next competitor, and will continue to have a superior (read safer) solution with the ability to operate autonomously on a higher percentage of streets for some time thereafter. The value of such a lead is quite big, I don't think it's fully priced in at this moment.
All that said it's entirely possible I've missed something, I've never personally worked on an autonomous vehicle, if @electracity, @myusername or @Value Ev or others believe Tesla will run into some problem that will take 10 years to solve, then tell us what that problem is.
I do think autonomy is the most important technology for the future of the automobile, I also think it's clear that Tesla has a significant lead here wrt existing products. I also believe, but can't prove, that Tesla has the best team and is making the fastest progress toward autonomy; I can't prove this, but I think the following are good reasons to believe it:
1) Their existing system is the best, despite having far fewer hardware resources than some other systems.
2) Their existing system is the best despite the fact that other companies have been working on lane-keeping systems for much longer than Tesla. For example Nissan first introduced lane keeping in 2001, Toyota in 2002, Honda in 2003.
3) Tesla has, in general, made it a higher priority than other manufacturers.
4) No other auto company has equal access to software talent due to Tesla's proximity to Silicon Valley and Elon's star power.
In addition to having the best system and making the fastest progress, Tesla's ability to issue OTA updates gives them a 1-2 year advantage over competitors as I think it's clear that the hardware is easier than the software (sure, you could have the dealership update firmware, but this is pretty kludgy and will become unwieldy when new safety related fixes are constantly required). It may actually be the case that realtime OTA updates are *required* for *any* autonomous vehicle in the future as it may be that a current database of all regulatory traffic signs/signals will be required in some countries for all autonomous vehicles.
Anyway I think understanding the tech involved will help people understand exactly how plausible full autonomy is at the present time. It's going to be a big day when the first transfer of liability from owner to manufacturer takes place. Below is an update of my first post on the topic from the ST thread, which I wrote hastily while in a meeting.
...Full autonomy is coming, I think in 5 years or less, probably less. That doesn't mean full autonomy on every road on planet earth, or on every road in the United States, it only means full autonomy on some usefully large subset of roads, probably starting with US interstates. I think the following technologies in combination, are sufficient for full autonomy.
1) GPS with SBAS (and soon LBAS) reliably gives accuracy greater than 12 inches, I see this accuracy every week, I fly aircraft that have it. This in combination with high precision maps will give the vehicle great situational awareness with regards to position on the network of streets. Vehicles will have redundant GPS systems.
Tesla actually has the ability to create their own LBAS system by using preset points on roads every 20 miles or so that have a precisely known location and that are identified by the camera system, I'd guess the camera system will have accuracy to a couple inches which means that the GPS signal can be corrected to within a couple inches.
2) Redundant binocular (+) camera systems will paint a 3D picture of the surroundings, recognizing lanes, vehicles, pedestrians, animals, street signs etc. This is the hardest part, machine object recognition has only recently surpassed that of humans.
3) Redundant radar systems will paint a 3D picture of the world that complements that of the binocular (+) camera systems. They will also see through some objects that the binocular (+) camera systems cannot and will give some density information allowing the system to distinguish a plastic bag blowing in the wind from a truck tire bouncing towards your face.
4) Ultrasonic sensors give a tertiary system for verifying objects that are near the vehicle.
5) Inertial navigation system, off the shelf GPS systems only update 10x per second you can get quicker systems but they're expensive. The inertial system will smooth GPS data and fill in for it in areas with poor GPS reception. These systems can have very high sampling rates, over 1000Hz.
6) Large database of relevant data including a complete roadmap, traffic signs/signals, obstacles, lane width, number of lanes etc.
7) Massive processing power: nVidia's PX2.0 has almost exactly 1000x the processing power of the Tegra 3 currently powering AP 1.0. They will need to have Redundant processing systems. [edit] User MP3Mike says autopilot does not use the Tegra 3, this may be the case, I believe it has more processing power than any other onboard chip but perhaps not.
It seems to me that ALL the technologies required to make a vehicle autonomous already exist. They just need to be combined with the right logic, I don't think this will be as difficult as some software problems that have already been solved.
If the above is correct, or in the vicinity of correct, it has serious implications for the value of Tesla. We've all seen the speculation about how the economics of personal transportation will change, given autonomy, from Adam Jonas and others. I suspect Tesla will get an autonomous capable product to market 3 or more years before the next competitor, and will continue to have a superior (read safer) solution with the ability to operate autonomously on a higher percentage of streets for some time thereafter. The value of such a lead is quite big, I don't think it's fully priced in at this moment.
All that said it's entirely possible I've missed something, I've never personally worked on an autonomous vehicle, if @electracity, @myusername or @Value Ev or others believe Tesla will run into some problem that will take 10 years to solve, then tell us what that problem is.
