Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla Auto Pilot Within Three Years

This site may earn commission on affiliate links.
I think the main challenge will be in crappy weather conditions. I was driving today in deep fog that had a total whiteout ca 30-40m from the car I assume (was able to see ~1.5 spans of the street lamp posts). That's extremely hard to decipher using a camera based system unless it has also filters and sensitivity to IR etc. Second crappy weather would be serious rain with splatter and dirt getting onto the cameras making the pictures blurry and out of focus in varying regions of the view. In good daylight conditions it's not that hard, but it's the crappy weather that has to be worked on...
 
I am still very sceptical about the whole concept. If Elon would prove me and the other sceptics wrong my stock would be happy so no problem here. I don't have to be in the right. But it is imperative that an autopilot system
is 100% fool proof. Saying it will apply to 90% of all situations is probably not going to cut it. Who decides about those 10% ? How are they handled. You have to take into account human behaviour at the moment You go down
to 99% fro 100%. That opens the door wide for dubious argumentation in case anything serious happens. If You do it this way it's bound to be a sue-fest.

Also I see a big problem in the interface, i.e. the point where the driver takes over or the other way around, where he gives control to the car. When will this happen ? I can easily imagine situations, where things get
a little narrow or hectic and the human driver suddenly loses trust in the system, takes over and causes an accident doing so.

I still see a Pandora Box of problems here, as much as I like Tesla to succeed.
 
Also I see a big problem in the interface, i.e. the point where the driver takes over or the other way around, where he gives control to the car. When will this happen ?

I'd suggest that the autopilot would alert when the conditions were beyond its capability. Something like this:

CURRENT MODE: Autopilot off
No indicators.

CURRENT MODE: Autopilot on standby
Conditions within autopilot capacity. Autopilot ready indicator green
Conditions not within autopilot capacity or marginal. Autopilot ready indicator red
In standby mode it would operate similar to Mobil Eye where it alerted the driver but would take no independent action.

CURRENT MODE: Autopilot on
Conditions within autopilot capacity. Autopilot engaged indicator green
Conditions marginal but within autopilot capacity. Autopilot engaged indicator yellow and audio alert (Driver can choose to leave it on or not)
Conditions not within autopilot capacity. Autopilot engaged indicator red and audio alert (Autopilot turns off)

The first release would have only limited times when the autopilot could be engaged. Perhaps partly GPS based. So rather than trying to figure out if it was in a parking lot by sensor readings it would do it by location as well as sensors. On standby, it would still give the driver clues.
 
I think Elon is thinking outside the box, like he usually does.

Remember, before Tesla, engineers would have laughed at the idea of building an EV out of laptop batteries. (Yes, i know these are not exactly laptop batteries, but you get the point). The industry standard was using industrial-grade, relativley low capacity but highly reliable components and chemistry, which, of course, also meant range was low, yet prices were high. Then came Elon, and he changed the paradigm.

With self-driving cars, one of the biggest issues is that they are trying to make them work in every single situation, every weather condition. Even Airbus and Boeing don't do that, AFAIK pilots only engage autopilot in good weather, at cruising altitude, for things like hours of flying in a straight line. This Google-approach means, they have to use lasers, high performance computers, etc. which make the car look and cost as much as an aircraft carrier.

I bet Tesla's system will warn and disengage when there is too much fog, too heavy rain or snow, ice, unexpected construction work on the road etc. and only work in "normal" conditions. On the flip side it will be simple and (relativley) inexpensive. A also think it is only intended for the next gen Model S, not the Model E, as it will still be a costly extra. I could be wrong about that, though.
 
I think Elon is thinking outside the box, like he usually does.

Remember, before Tesla, engineers would have laughed at the idea of building an EV out of laptop batteries. (Yes, i know these are not exactly laptop batteries, but you get the point). The industry standard was using industrial-grade, relativley low capacity but highly reliable components and chemistry, which, of course, also meant range was low, yet prices were high. Then came Elon, and he changed the paradigm.

With self-driving cars, one of the biggest issues is that they are trying to make them work in every single situation, every weather condition. Even Airbus and Boeing don't do that, AFAIK pilots only engage autopilot in good weather, at cruising altitude, for things like hours of flying in a straight line. This Google-approach means, they have to use lasers, high performance computers, etc. which make the car look and cost as much as an aircraft carrier.

I bet Tesla's system will warn and disengage when there is too much fog, too heavy rain or snow, ice, unexpected construction work on the road etc. and only work in "normal" conditions. On the flip side it will be simple and (relativley) inexpensive. A also think it is only intended for the next gen Model S, not the Model E, as it will still be a costly extra. I could be wrong about that, though.

Agree. Like anything else it will be done in phases and will improve as each generation of self-driving capabilities is available.
 
I am still very sceptical about the whole concept. If Elon would prove me and the other sceptics wrong my stock would be happy so no problem here. I don't have to be in the right. But it is imperative that an autopilot system
is 100% fool proof.

LOL! The guy can start with spare parts and build the best Space/Rocket program in the world. He can put satellites in orbit and dock with a tiny station in space. I think he can safely get a car to grandma's house....
 
First responses to Causalien's initial nuts-&-bolts post:

1. License plates do NOT uniformly have a white background, so that's a very minor tweak.

2. Although I like the "cars move fast so there's one algorithm to deal with them; people are essentially immobile so another algorithm takes care of them" argument, now you have to throw into the mix:
*motorcycles - small, fast and with a reputation, deserved or undeserved, for jinking. And with far smaller license plates, too
*bicycles - small, slower than most cars but not so immobile as pedestrians, and with a reputation, deserved or undeserved, for jinking. And with infnitely smaller license plates, too. On the other hand, their tires, etc., might become the default sizing=distance determinant
*joggers/sprinters - pedestrian-sized objects with a reputation for jinking
*and then there are dogs, cats....or where I live, caribou and moose (talk about immobile!)
 
I am still very sceptical about the whole concept. If Elon would prove me and the other sceptics wrong my stock would be happy so no problem here. I don't have to be in the right. But it is imperative that an autopilot system
is 100% fool proof. Saying it will apply to 90% of all situations is probably not going to cut it. Who decides about those 10% ? How are they handled. You have to take into account human behaviour at the moment You go down
to 99% fro 100%. That opens the door wide for dubious argumentation in case anything serious happens. If You do it this way it's bound to be a sue-fest.

Also I see a big problem in the interface, i.e. the point where the driver takes over or the other way around, where he gives control to the car. When will this happen ? I can easily imagine situations, where things get
a little narrow or hectic and the human driver suddenly loses trust in the system, takes over and causes an accident doing so.

I still see a Pandora Box of problems here, as much as I like Tesla to succeed.

Do you use cruise control? To me it's absolutely no different in principle and in fact auto-pilot would be better than current dumb cruise control systems that are found in the vast majority of cars. Why are you worrying about this, when you should be railing against the millions of accidents waiting to happen right now?

Auto-pilot would include the adaptive cruise that's found in cars with advanced safety systems (and is by far the most popular component). The car has a set of parameters within which it is expected to operate and will run on auto-pilot as long as it remains within those parameters. Auto-pilot's parameters will never ever be outside human norms, because otherwise the car would get into situations where it would be dangerous for the driver to take over.

The only question is how much software will be able to do.

What I really like about Tesla's approach is that they're starting with the end goal in mind. The end goal defines the hardware requirements, which should mean that they'll be able to sell systems (ha, cars) with the hardware and limited software capability and then upgrade the software to introduce new features. In the traditional model, manufacturers use technology to sell hardware upgrades, but the risk of that is that when you lag the competition you are forced to cut profits to compete. With the Tesla model you have a primary focus of leading on the hardware, with the knowledge that the software can catch up. This is part of the reason I always expect delays from Tesla.
 
My reasoning is this:
Even on highways and autoroutes, the primary failure case to avoid at all case, is hitting a pedestrian or a car. Because once you do hit one, a tsunami of politicians will vie to shutdown any type of autonomous driving. "Robot kills Human" will make for some great headlines and gain you more votes from your constituents.

Therefore, designing the AI to avoid collision with human and cars will have to be built right away. Just because you are on a highway, doesn't mean that you can stop detecting for collision with a human. So if that system is always active, then it means that you might as well design it to be able to work in cities and normal roads where pedestrians are present.

What I am getting from you guys is that $5000 is too much for this autopilot package?

Elon released that video of him playing with a leap motion controller and various display systems using high end Siemens CAD. The other components were quite expensive but the 3D motion tracking package with two cameras, infrared, and the processng chip (Leap Motion) is about $80 at Best Buy. Check it out. That provides real time precise tracking of all 10 fingers...an order of magnitude better performance than xbox connect...the other model.

For actual product I think he expects the cameras/sensors and hardware processing to be very inexpensive and unobtrusive, something like Leap Motion adapted for longer distance and deployed in a fixed 360 array. Maybe a dozen or more cameras with integral motion capture processing chips and infrared arrays together costing less than $1000.
 
Last edited: