Tesla autopilot has been around for many years. The autopilot aids drivers by using cameras, sensors, and sonars. Tesla’s autopilot was not always what it is now.
Tesla originally partnered with Israeli vision technology company, Mobileye for its driver assistance programs, including autopilot. After a split with Mobileye in 2016, Tesla worked with Nvidia for some of the AI tech, before transitioning to their own autopilot hardware in 2018.
Read on to learn more about the beginning stages of Tesla autopilot and how it has evolved over the years. Also, why certain decisions were made to continue upgrading the autopilot.
Who Made the First Version of Tesla Autopilot?
The first version of the Tesla autopilot was created in 1999 by Mobileye N.V. which was founded by Amnon Shashua and Ziv Aviram. Their company is in Israel. They went on to create an SoC processor with software to detect collision warnings and lane departures.
The technology known as Autopilot Hardware 1.0 was then placed into all Tesla’s Model S and Model X vehicles. These vehicles were built between September 2014 and October 2016. These cars were equipped with:
- One camera
- Radar
- Sonar sensors
And how did it work?
How Autopilot Hardware 1.0 Worked
The vehicle’s computer was processed by the Mobileye EyeQ3 processor. Through the camera, Mobileye’s chip helped vehicles steer and stay in their lanes. At no point whatsoever is the autopilot supposed to take the place of a human driver. The drivers of these vehicles need to always be paying attention.
The one camera that is a part of Tesla’s Autopilot Hardware 1.0 runs all the information it sees through the processor created by Mobileye to help aid the autopilot.
Mobileye is recognized as a global leader in vision technology and the development of Advanced Driver Assistance Systems (ADAS) and autonomous driving. While the program processed by Mobileye and Tesla are not completely autonomous, the autopilot falls into this category.
The First Autopilot Worked with the Driver, Not in Place of the Driver
The software Mobileye created that can detect collision warnings and lane departures that allows the vehicle while in autopilot to let the driver know something can happen or something needs to be avoided.
The part that Tesla had added in is that the drivers are made aware when autopilot is engaged, that the driver needs to always be paying attention to the road and their hands need to stay on the steering wheel. The Tesla autopilot support reminds drivers to keep hands on the steering wheel at all times and stresses that drivers must maintain control and responsibility for the vehicle.
The autopilot available during the release of hardware 1.0 was the beginning stages of Tesla developing a self-driving car. It was the very first version of what would eventually become the autopilot that is seen in Tesla cars today.
But the partnership didn’t last.
Why Do Mobileye and Tesla No Longer Work Together?
The partnership between Mobileye and Tesla ended after a fatal car crash in May 2016 involving a Tesla Model S with autopilot on occurred. Neither the driver nor the autopilot on the vehicle hit the breaks or started to slow down when a truck changed lanes in front of the Tesla. The camera for the autopilot did not recognize nor did it see the truck cross over into the lane.
It was Mobileye who ended the partnership, stating that Tesla was “pushing the limits” when it came to safety. In an Entrepreneur magazine article, the Chief Technology Officer for Mobileye expressed that Hardware 1.0 was intended to be an assistance system, not a driverless system.
Tesla responded by updating Autopilot to make it more difficult for drivers to avoid safety warnings. And it stood behind its assertion that assisted driving is still significantly safer than purely manual driving.
Changes to Tesla Autopilot After Fatal Florida Accident
On September 23, 2016 version 8 of the Tesla autopilot had changes made. One feature of the autopilot is Autosteer which allows vehicles to steer, accelerate, or brake automatically. The changes made to Autosteer were:
- Additional warnings. Drivers would be alerted three different times to keep their hands on the steering wheel, otherwise the Autosteer would deactivate for the next hour while the car was still in use.
- 10-minute warning. When a vehicle is traveling 8 mph-45 mph the alert will display 10 minutes after having hands off steering.
- 3-minute warning. Any speed over 45mph the alert will display three minutes after hands off steering.
Adding these extra precautions in can help drivers that are using the Autosteer feature to pay attention more if that is something that is needed to be done of them. Having the alerts show up more often or within a shorter amount of time when the vehicle is traveling at a higher speed can also make drivers more aware of their surroundings.
A New Age in Highway Safety
On January 19, 2017 the National Highway Traffic Safety Administration had their office of Defects investigate the accident as well. There were no defects found in the design or performance of the autopilot for the Tesla involved in the accident.
The 63-page incident report includes several pages on regulation and policy regarding automated vehicles. A new consideration that would prove to grow to be a necessity over the coming years.
Transitioning from Mobileye to Tesla Hardware
In 2017 Mobileye was purchased by Intel for $15 billion dollars. Being purchased by Intel, the deal that Mobileye had with Tesla ended once their original contract was over. Mobileye said in a statement that it would not perform further hardware updates for Tesla but would support ongoing upgrades to Autopilot.
So, any of the Tesla Model S or Tesla Model X vehicles that were made with the autopilot hardware created by Mobileye would still be upgraded as needed until a certain point. As far as the hardware that Mobileye had created for Tesla, that would no longer continue.
Tesla in a statement said only that it was “transitioning to internally developed software for the camera portion of Autopilot.”
Once Mobileye discontinued working with Tesla, the company itself either already was or was planning on internally producing the software needed that would take the place of Mobileye.
After the partnership with Mobileye ended, Tesla began working on doing more of their work in house because over a period of time released three more versions of the autopilot, with the newest being available today. A lot of the new features that were added to the autopilot after the partnership with Mobileye had to do with what came out of the accident report.
Tesla Autopilot Hardware 2.0 and Hardware 2.5
The autopilot Hardware 2.0 takes a giant step up from what was available for Hardware 1.0. The new hardware started appearing in cars built between October 2016 to August 2017. Hardware 2.0 came equipped with
- Eight cameras
- Same amount of radar
- Sonar sensors with upgraded range
The SoC chip available for Hardware 2.0 was created by Nvidia. Nvidia supplied Tesla with the hardware 2.0 computer that started getting installed in all Teslas produced from October 2016. Three of the cameras available are in or around the same area that Mobileye had their front facing camera in the autopilot with hardware 1.0 in it.
The Addition of 7 More Cameras
The added camera angels were most likely put in place due to the fatal accident that had occurred with the Model S that was on autopilot. Having more cameras able to view more space around the front of the car would give the vehicle the ability to see a car switching lanes easier.
New Capabilities
One driver posted a video on his YouTube channel showing what the autopilot equipped with the hardware 2.0 is capable of. He starts his journey from his driveway to his destination. Along with views from the many different camera angels available the car is seen:
- Stopping at stop signs
- Stopping at red lights
- Merging and taking exit ramps
- Changing lanes
- Once Daniel gets to his destination he is seen getting out of the car. The car then, all by itself:
- Drives off to find a parking spot
- Comes to a complete stop for a pedestrian to cross
- Self-parks in a parallel parking spot
While the hardware 2.0 came out around the same time the hardware 1.0 was done being used. The hardware 2.0 was not available for as long as hardware 1.0 was. Hardware 2.0 was available less than a year before it got an upgrade.
Tesla Autopilot Hardware 2.5 – With Nvidia Components
The Tesla Hardware 2.5 is like Hardware 2.0. Hardware 2.5 is more of a revision to hardware 2.0 in that:
- The radar supplier changed to Continental, which gave it a farther radar range.
- There are two Nvidia SoC chips instead of just one.
The 2.5 version of the Tesla hardware was built in vehicles from August 2017 to March 2019. Vehicles with the 2.0 hardware had a shorter availability time before 2.5 hardware was released.
Since Nvidia was already a part of the 2.0 hardware, having a second chip put in for the 2.5 version would not have been hard. The radar supplier changing is probably the main reason for the 2.5 upgrade so there was a way to differentiate between the two types of hardware.
Tesla Autopilot Hardware 3.0 – Transitioning to Full Tesla Production
Tesla hardware 3.0 is also referred to as full self-driving and was released in vehicles starting in 2019.
Because development of the chip and hardware was mostly complete and the new 3.0 hardware was basically just refinement and improvements, the new hardware cost significantly less than version 2.5.
One the partnership between Mobileye and Tesla became nonexistent, Tesla began working on developing their own ways of building what was needed to run their cars. Nvidia was still a part of hardware 2.0 and 2.5, but once the full self-driving was released it was all done by Tesla.
The Promise of Full Self-Driving Mode
Since the release of hardware 2.0, Tesla has been offering their customers the ability to purchase the full self-driving upgrade for when it finally became available.
Even though it is considered a purchasable upgrade, it did not have to be bought up front right away. The FSD upgrade would become available at a higher price if it were purchased after the vehicle was purchased.
Tesla has created hardware versions with different names:
- Enhanced autopilot (EAP)
- Autopilot (AP)
- Full Self Driving (FSD).
Finding out if vehicle is available to get the upgrade can all be done from the touchscreen within the car. If a Tesla vehicle has the computer with hardware 2.0 or 2.5, that person will be able to purchase the FSD.
- Samsung is now creating the FSD chip for Tesla that is available for hardware 3.0. The camera image processing that is available for this hardware is more powerful which in the end runs better due to the artificial intelligence hardware that Tesla has created.
- Tesla Raj goes over the newest Tesla software that was released in 2019 which is all a part of hardware 3.0. One important new feature is the dashcam can now save image clips when the horn is honked. This way if the driver is ever in an accident or something happens that cannot be avoided, the images taken can help show how something ended up happening.
Collecting Data for Future Improvements
The amount of information that the cameras view outside and inside the car all gets saved online. This information is then processed along with anyone that is following the Tesla community to share more information or let it be known about anything that can be fixed or could possibly be changed.
Who Is Behind the Autopilot Hardware Today?
There are many people working on the Tesla autopilot today. The three people a part of the autopilot team specifically are:
- Andrej Karpathy – Director, AI for autopilot
- Peter Bannon – VP, Hardware design engineering
- Stuart Bowers – VP, Engineering
Andrej Karpathy started working for Tesla in 2017 as the director of artificial intelligence, eventually moving up to the Senior Director of artificial intelligence. Tesla uses the artificial intelligence mixed with their cameras to be able to pool the data and figure out how to go from there. What needs to be upgraded or what is best where it is. This way too, any maps that are available to be used are not a part of the mapping capabilities available in Tesla vehicles.
Peter Bannon started working for Tesla in 2016 as the VP in hardware design engineering. Being a hardware design engineer for the autopilot team he helped Tesla work on their current hardware available and began helping to work on the hardware that would be releasing as the full self-driving available in the upcoming and newer Tesla vehicles.
Stuart Bowers worked for Tesla for a little over a year. He worked on the autopilot software that was put into the vehicles. He started in 2018, so he also would have been a part of the team that would have been working on the newest hardware available for Tesla.
The processing power of the autopilot along with the full self-driving being made available in the newer Tesla vehicles today has come a long way from where it once started.
Can Your Tesla Autopilot Be Upgraded to Hardware 3.0?
Black Tesla has an informative video that breaks down all the different hardware versions that have been available for Tesla leading up to Tesla reaching out to their customers about upgrading their vehicles to hardware 3.0. The following is a quick breakdown of the information in the video.
Media Control Unit (MCU) is a part of the Tesla autopilot.
- MCU 1 is available in Tesla vehicles built from 2012 to February 2018.
- MCU 2 is available in Tesla vehicles built from March 2018 to the present
Electronic Control Unit (ECU) -the different variations of the hardware, 1.0 to 3.0:
Hardware 1 | Only pairs with MCU 1.Cannot get hardware 3.0 due to the processing power of the computer not being able to handle what is available in 3.0.This version only has the one forward facing camera. |
Hardware 2.0 | Paired with MCU 1.Does have extra cameras compared to what is available for hardware 1.0.Still cannot support the processing power for the 3.0 due to there being an MCU 1. |
Hardware 2.5 | Cars bought that were built prior to March 2018 have an MCU1 installed: These cars will not be able to be upgraded to the 3.0 because there is no upgrade available from Tesla to upgrade from MCU1 to MCU2.Tesla would have to create this upgrade themselves for you to get the 3.0 upgrade. Cars bought after March 2018 with this hardware will be able to upgrade to the 3.0. The cars offering the MCU2 have the processing power to handle version 3.0. |
Based on when this video was made, customers that want to upgrade to the 3.0 have had to have already purchased the full self-driving capabilities for their vehicles.
Conclusion
All in all, the autopilot for Tesla has been produced by different companies. The multiple different versions of hardware have increased the capabilities of the autopilot over the years. Tesla started out working in partnerships and ended up designing much of the autopilot in house.