Tesla putting hardware for full autonomy in all models; temporary loss of some Gen1 Autopilot functions
20 October 2016
Tesla announced that effective immediately, new Tesla vehicles—including Model 3—will have the hardware needed to support full autonomous driving.
The required software for full autonomous driving is still under development and will need validation and regulatory approval. In fact, Teslas with the new hardware will temporarily lack certain features currently available on Teslas with first-generation Autopilot hardware, including some standard safety features such as automatic emergency braking, collision warning, lane holding and active cruise control.
As Tesla proceeds with its testing and validation, it will enable those missing features over the air, together with an expanding set of new features, it said.
Tesla’s hardware system includes:
8 surround cameras (up from four) providing 360 degree visibility around the car at up to 250 meters of range;
12 updated ultrasonic sensors, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system; and
a forward-facing radar with enhanced processing to provide additional data about the world on a redundant wavelength, capable of seeing through heavy rain, fog, dust and even the car ahead.
The vehicles will feature a new onboard computer with more than 40 times the computing power of the previous generation to run the new Tesla-developed neural net for vision, sonar and radar processing software.
Model S and Model X vehicles with this new hardware are already in production.
Tesla notes that self-driving functionality is dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction. Tesla also cautioned that using a self-driving Tesla for car sharing and ride hailing for friends and family is fine, but doing so for revenue purposes will only be permissible on the Tesla Network, details of which will be released next year.
Enhanced Autopilot. An enhanced version of Autopilotwill match speed to traffic conditions; keep within a lane; automatically change lanes without requiring driver input; transition from one freeway to another; exit the freeway when your destination is near; self-park when near a parking spot; and be summoned to and from the driver’s garage.
Tesla expects its Enhanced Autopilot software to complete validation and be rolled out via an over-the-air update in December 2016, subject to regulatory approval.
So the Tesla's are to in practise have not even the basic assisted driving features found in common or garden cars, although buyers will have paid for the hardware.
Posted by: Davemart | 20 October 2016 at 12:35 AM
Inside EVs says:
'Tesla’s website/order configurator has already been updated, and “Enhanced Autopilot” adds $5,000 to the cost ($6,000 after delivery) – up from $3,000 earlier. And the promise of “Full Self-Driving Capability” in the future (without a date due to regulatory issues) adds another $3,000 ($4,000 after delivery).'
http://insideevs.com/all-tesla-vehicles-in-production-are-equipped-with-self-driving-hardware/
which on the face of it contradicts that the hardware is to be fitted to all models, unless they are referring to the activation charge, which is not happening yet anyway
Posted by: Davemart | 20 October 2016 at 12:42 AM
This is epic good news. Getting the needed hardware for full autonomy cars into production cars happened 3 years earlier than I expected it to happen for Tesla (I thought 2019). I always thought Tesla would be first with fully autonomous cars but that they do it today for order today is just blowing my mind. The true BEV revolution starts now. Now that subsequent software update will make all of Tesla’s cars produced from today and forward fully autonomous Tesla’s software engineers will solve the three remaining issues with BEVs:
1) Cost because Tesla cars can now do 100k miles per year and thus take full advantage of their lower marginal driving cost and use that to pay for higher upfront cost. When you do not use the car yourself others can use it as a self-driving taxi.
2) Range because you can always jump into another self driving taxi and carry on.
3) Charging time because again you can just jump into another self driving taxi and carry on.
Also super nice to see that Model 3 will also be a fully autonomous car and that Tesla is planning to make 500,000 fully autonomous cars in 2018 and going up to 1 million fully autonomous cars by 2020. Remember that 1 million self-driving BEVs can do the transportation services of 6 to 7 million non-self driving cars so Tesla probably still being the only self driving taxi company in the world by 2020 (old auto industry CEOs are still rambling about fully self-driving cars are not possible until 2025) will be very close to overtake Toyota, GM and VW as the world’s largest automaker. All Tesla need is to make 2 million fully autonomous cars per years something that would happen by 2022 and they will have the equivalent production of 13 million non-self driving cars per year.
The Nevada Giga factory by the way will be upgraded to a 150Gwh factory. Assuming all goes into model 3 and that Model 3 will use a 55kwh battery that is enough for making 2.7 million such Model 3 cars per year. That giga factory is also going to make backup batteries and larger battery packs for Model S and X and other car and truck models so that will not happen. However, Tesla is building the capacity today to become the world largest auto maker by 2022 and this is nice to know.
BEVs will no longer be niche and Tesla could transform the entire global auto industry into a fully sustainable BEV only making industry all by themselves if needed. Tesla does not wait for anyone to get along. Either you step it up and embrace fully self-driving BEVs or you go bankrupt by 2025 or before that.
Posted by: Account Deleted | 20 October 2016 at 12:48 AM
Apparently AEB, front and side collision warning and auto high beam are due to be enabled in Dec 2016:
https://www.tesla.com/autopilot?redirect=no
Posted by: Davemart | 20 October 2016 at 12:49 AM
You are such a jerk Davemart and you seem to never learn anything. You should try self improvement therapy. You need it.
Posted by: Account Deleted | 20 October 2016 at 12:56 AM
Getting the hardware in is one thing. The hardware is easy. The software is the problem.
He is clever to release the h/w early so he can get lots and lots of training data and experience in what can go wrong.
Then they have to get regulatory approval, which may take longer again.
Then, they will take further regulatory approval to be used as autonomous taxis, and in particular for unmanned cars driving around cities to position themselves.
I wouldn't hold my breath on that one.
Getting cars into the hands of individual owners where one person is responsible for them is one thing, having them roaming around cities unmanned is quite another.
(I am sure it will happen, it will just take longer than Henrik thinks).
Posted by: mahonj | 20 October 2016 at 01:06 AM
The reporting on this news about Tesla is incorrect. The Gen 2 hardware suite has 8 cameras whereas the Gen1 hardware suite only had one camera (not 4 as reported).
The 4 cameras is the cameras that are activated when Tesla buyers only opt for the Enhanced Autopilot costing 5000 USD, customers buying the full autonomy option (costing 8000 USD) will have all 8 cameras activated. All cars including Model 3 will have the full autonomy suite build in whether you buy it or not. This can always be bought and activated at a later point in time. It is worth noting that the one camera in Gen1 was made by Mobileye whereas the new 8 cameras are developed and produced by Tesla.
Posted by: Account Deleted | 20 October 2016 at 01:18 AM
Mahonj I was too pessimistic about when fully self-driving hardware would enter production cars by 3 years!
Now with this surprising news I think that Tesla will be allowed somewhere on the planet and perhaps limited to preapproved roads to operate their cars as self-driving taxis sometime by 2018 and I think it will expand rapidly from there and cover most of the world by 2020 for Tesla’s cars. I am curious when exactly you expect it to evolve?
Also Tesla can test and improve their self-driving hardware in all Tesla cars even when the Autopilot is not activated. It still runs in the background communicating with Tesla’s datacenter and making up to date maps and learning from the driving behavior of the human drivers. In just 12 months from now Tesla will get data from possibly over 130.000 Tesla cars that are equipped with this hardware on a daily basis. It will not take long to get the needed data to document that this is working as intended and will save lives on a massive scale when allowed. The regulators will be quick to adopt this. Who will be against saving 1.2 million people from dying in traffic accidents every year on this planet? This is one of the few cases where everybody wants it to happen ASAP.
Posted by: Account Deleted | 20 October 2016 at 02:33 AM
Henrik:
It is such a shame that in the absence of anything worthwhile to say, you continue to post.
I am afraid that others do not share your opinion that Musk is a God, and so entertain criticism of him, however much that offends you.
In the finest stroke of genius since the Falcon wing door, Tesla have produced a substantial number of cars for inventory for Europe.
Old hardware.
So they will have to unload them, aside from paying to install non-functioning hardware in their new cars.
That is not great for making a profit, were Tesla to have any such ambition.
Of course their actual business is issuing shares, which they are very good at.
Posted by: Davemart | 20 October 2016 at 02:44 AM
At Tesla’s news blog about this upcoming fully autonomous Autopilot there is a video demonstrating this new hardware with a test version of the upcoming software. See link below.
It shows that the car can handle stop lights and intersections. It even drives around without a driver at a Tesla parking lot to find a free parking spot. When it does it parks flawlessly. And it stops automatically for a person at the parking lot that block its way temporarily. It is very impressive. They are obviously very far in terms of making Tesla cars drive everywhere on their own.
https://www.tesla.com/blog/all-tesla-cars-being-produced-now-have-full-self-driving-hardware
Posted by: Account Deleted | 20 October 2016 at 03:09 AM
I rarely comment on Tesla threads here as I am not an enthusiast, and have no intention of trolling every Tesla thread as some do on hydrogen and fuel cell ones.
This is so egregious however that I was moved to comment.
They have nothing working, so are sticking in the hardware and hope to get something working at some time in the future.
It does however serve to lay bare that their previous hardware was in no way adequate, as I said when it was introduced, and as it and its associated software killing people tends to confirm.
Posted by: Davemart | 20 October 2016 at 03:37 AM
Henrik,
Read the rules for this site.
Posted by: SJC | 20 October 2016 at 10:09 AM
Neither this story nor the Tesla press release defined the full extent of the "Full Self-Driving Hardware".
In addition to the sensor enhancements, does it imply that all vehicle systems are hardware-capable (if not yet enabled) for full autonomous ops? This would include full-authority forward/reverse motion, braking, turning, parking, signalling, lighting, lock/unlock, maybe even the horn.
With any luck, we'll learn of the full capabilities on Tesla's schedule, and not well before that due to hackers.
Posted by: Bob Niland | 20 October 2016 at 01:23 PM
It's one thing to handle a stop sign and a traffic light automatically at an intersection with all sorts of nice visible lane markings.
It's quite another to handle it when all the markings are covered in snow. Or glare ice. Or the intersection is under construction and there aren't any markings. Or the power is out and a police officer is directing traffic. Or (as happened in a major blackout in Toronto in 2003) everyday citizens were directing traffic because there weren't enough police officers to cover all the intersections and they had enough other headaches on their hands.
And even if it's sophisticated enough to handle that (which I doubt) ... How does the software distinguish a citizen directing traffic in the midst of a blackout ... from a teenager playing havoc, which you know is going to happen!
I have little doubt that Tesla will be able to produce a better "autopilot" that works in more circumstances and that those circumstances will constantly increase in scope over time, but they are a loooooong way from being able to be summoned autonomously in the midst of a power blackout due to a snowstorm.
Posted by: Brian P | 20 October 2016 at 05:38 PM
Auto-pilots, anti-collision secondary/proximity radars and improved satellite navigation systems have made flying a lot safer, even in dense trafic at high speed. The main reason why pilots still sit in the cockpit are to satisfy passengers expectation, pilots and air traffic controllers and their powerful unions.
Some/many ADVs will be produced and used by 2030 or so but essential effective software, anti-hacking and fail safe multi-sensors have to be greatly improved to make the system safer and secure in all conditions.
Wireless voice commands for your ADV may also become a low cost add-on option.
Posted by: HarveyD | 20 October 2016 at 07:16 PM
Bob all the hardware for the fully self-driving Autopilot is build into every vehicle that Tesla produces whether you order it or not. At Tesla’s configuration page (see link below) you can choose to buy and activate the Enhanced Autopilot hardware and software for 5000 USD at order time. If you do not order it and one year later regret it you can still buy and activate the feature but now it will cost you 6000 USD. You get a rebate for ordering faster. The full autonomous Autopilot will cost an additional 3000 USD and 4000 USD if you order it at a later point in time.
Brian the new version of Teslas Autopilot will be able to 3D map the environment and calculate markers in that environment that can be used for precision navigation (1 inch) without using GPS or lane markers. These maps are crowd sourced from all Tesla vehicles. They only requite one Tesla vehicle to drive on a road once and then all subsequent Teslas that drive through will know exactly where they drive within the inch even though the lane markers are not visible.
I read elsewhere that Musk plan to drive a Tesla from LA to NY on this fully self-driving Autopilot by the end of 2017 without any human intervention and neither any human intervention at the charger stations that will be upgraded with robotic chargers during 2017 and forward. Musk plan to start a nationwide autonomous taxi service sometime in 2018 for the US both using a fleet of Tesla owned vehicles and vehicles made available to that Tesla Network by private Tesla owners.
All around the world you bet that CEOs of the old automakers but also Apple, Google and Uber etc are holding crisis meeting about how to step up their own market introductions of self-driving taxis as they are far more behind Tesla than they used to think until this announcement from Tesla.
Harvey airpilots sit in the cockpit to fly whenever the Autopilot cannot just like Tesla’s limited Autopilot. A fully autonomous Autopilot for airplanes does not yet exist but it is getting close for that also. In 5 years time we will not need airplane pilots either.
https://www.tesla.com/models/design
Posted by: Account Deleted | 20 October 2016 at 11:37 PM
More they get paid, harder they will be to remove. It will be relatively easy with $10/hour taxis drivers but nine times harder with $90/hour ATC and 27 times harder with $270/hour captains.
Look how difficult it was to remove 'firemen' from diesel/electric locomotives. A very high percentage of train accidents are created by current drivers.
Fully automated road/rail vehicles could check, record and report road/rail conditions at very little cost while human drivers prefer to overlook.
Posted by: HarveyD | 21 October 2016 at 02:01 PM
Airplane pilots have the ability to "think out of the box" when something goes wrong. They have the ability to judge whether a situation warrants emergency actions or can safely be ignored. They can manually pop the circuit breaker on a faulty electrical circuit. They can judge whether they can safely land with one of the landing gears not fully engaged.
A fully automated aircraft without such an experienced pilot on board ... crashes.
Posted by: Brian P | 21 October 2016 at 04:40 PM
Brian apart from rewiring the circuit breaker in flight there is nothing a well trained AI autopilot cannot do that a human can do and the day will come where it can do more than a human brain without augmented synthetic AI build in. People still don't get that when you program and organise a computer like a human brain it will function like a human brain just a million times faster.
Posted by: Account Deleted | 22 October 2016 at 12:16 AM
Somewhere in between Davemart's pessimism and Henrik's optimism lies a likely scenario w/o the personal affronts please. I like the potential learnings that the inactive hardware allows, and that information provides a host of test scenarios for any new s/w development and any potential AI maturity. I doubt if the volume cost of the OEM hardware on a volume scale negates the efficiency of uniformity in production by much. It doesn't matter much now but the future benefits are huge. Just don't enable the full autonomy until it is appropriate.
Posted by: william g irwin | 22 October 2016 at 04:04 AM
Henrik: "...all the hardware for the fully self-driving Autopilot is build into every vehicle that Tesla produces whether you order it or not."
Yes, I now see reported elsewhere that all the needed servos are pre-installed, and also that Tesla says turning them on is simply an over-the-air download (assuming the internet is actually working at that moment, I suppose ☺).
Posted by: Bob Niland | 22 October 2016 at 09:54 AM
By 2025/2030 anti-ADVs posters may/will have growing problems try to buy a new vehicle without partial or full ADV facilities built in.
Shared mini e-buses with controlled ADV facilities will transport 6 to 10 passengers (on demand) from home to work places and/or main subway and e-bus stations and replace 6 to 10 existing ICEVs, reducing both GHG and traffic jams.
Driven ICEVs could be banned in many city centers after 2030 or have to pay for costly operation permits etc.
Henrik may be a few years ahead of the time but he is right with ADVs and other future automated tasks.
Posted by: HarveyD | 22 October 2016 at 10:01 AM
Just to be clear, I'm not opposed to the idea, I think it has its applications, I just have no interest right now to have or use such a vehicle myself. Maybe when I'm 90 years old (that's one of the applications). Elements of this technology have a use in preventing people from getting in trouble (e.g. preventing someone from driving through a red light or stop sign with other traffic coming, or preventing someone from making a left turn in front of another driver).
I think the average person is a lot more opposed to the use of public transit than the optimists think they are. Car-sharing is likely to get people OUT of public transit and it won't help reduce congestion.
And for all those idealists who think everything in the future is going to be automated ... I'd like to know whether those people think their own jobs are going to be automated. I know that automation in the world of manufacturing has replaced drudgery (an army of people each holding a manual spot-weld gun and pushing a button a thousand times per shift) with other jobs requiring skill - someone has to design the tooling, program the robot, and fix it when it breaks (and I'm one of those people). And someone still has to handle the parts that are going into the automation equipment ... It isn't practical to automate everything, and what would people do for a living if we did?
Posted by: Brian P | 22 October 2016 at 01:02 PM
I think Tesla’s rollout plan for the autonomous software is pretty simple. When their engineers and internal testers judge that a feature of the system (like adaptive cruise control with fully automatic lane change, (does not exist yet)) is ready for deployment Tesla will apply for permission to deploy that feature jurisdiction by jurisdiction. When they get that permission they deploy it. Eventually the number of features added to the system will make it a fully autonomous Autopilot. At that point Tesla must wait until they gather enough real world accident data in their datacenter monitoring all Tesla vehicles to conclude without any doubt that driving with Autopilot activated result in 50% less accidents >in all driving conditions< as when humans are in control. With this evidence on the book Tesla can apply jurisdiction by jurisdiction to be allowed to operate Tesla cars without human drivers behind the wheel and also apply for permission to start making versions of their vehicles that skip all the human controls like steering wheel, gas pedal and braking pedal. The latter vehicles will be operated mostly by the Tesla Network as taxis and owned by Tesla. When Tesla gets the permission they start deploying their Tesla Network jurisdiction by jurisdiction. Also Tesla owners will finally be allowed to drive hands free for as long as they wish or hail their cars from anywhere without any people in it. Finally, Tesla should also be allowed to sell versions of their cars without human controls to people that do not have a driver license like old people and disabled.
I expect Tesla will get the fully autonomous system ready by late 2017 or early 2018 and the first permits to use the cars as fully autonomous cars will come in some jurisdictions shortly thereafter (one or two months). I hope the US will allow Tesla to start the Tesla Network in all of the USA by the end of 2018 and that Tesla will get permissions to drive everywhere it is sold globally by the end of 2019.
Tesla will also start making other vehicles that will all be fully autonomous from launch. Like Model Y (the Model 3 with falcon dors), the minibus (a bus version of Model X for municipal transportation services that replace polluting diesel busses) the semi that will do away with heavy duty diesel trucks. A roadster and a pickup truck. All these vehicles will be launched before 2020.
It is worth noting that Tesla could save 600.000 humans from dying each year in traffic accidents if their first version of their fully autonomous autopilot was deployed globally to all vehicles. Also recall that Tesla’s end goal is to make a version of Tesla’s autopilot that eliminates 90% of all accidents so that 1.080.000 lives could be saved per year globally in traffic accidents. Not deploying these systems is deliberate mass murder and people that try to stop the deployment should therefore be sued for participating in mass murder.
Posted by: Account Deleted | 23 October 2016 at 01:15 AM
Deploying such systems prematurely, before the technology is ready for prime time (and it isn't), would be "mass murder".
It will eventually happen but it will take a lot longer than Henrik predicts it will.
Liability is one thing Henrik has not considered. When an individual driver causes a collision the liability repercussions are pretty limited to that driver's insurance coverage. When a big faceless corporation does something that causes an injury, the lawyers get dollar signs filling their eyes.
Posted by: Brian P | 23 October 2016 at 06:19 AM