Tesla’s Autopilot system is one of the most famous engineering systems on the automotive market. Few people understand how fast this technology has evolved since it was first conceived at Tesla back in the early 2010s.
The history of Tesla’s autopilot stretches back to 2013 when it was first teased by Elon Musk all the way to late 2020, when the first fully self-driving vehicles began beta-testing. While Tesla’s autopilot hardware and software isn’t fully self-driving yet, it gets closer every year.
There are several car companies on the market these days that are pushing self-driving technology. However, Tesla led the charge on this capability and they’re still at the forefront of innovating it. Keep reading to learn more about how Tesla’s autopilot program was started and where it’s headed in the future.
Tesla started considering autonomous cars in the early 2010s, a decade after the company was founded. In 2013, Elon Musk started kicking around the idea of autonomous vehicles, equating them to airplanes. Citing the safety and efficiency of using autopilot in airliners, Musk stated that he believed autopilot would be a good idea for cars, too. (Source: Yahoo Finance)
When Tesla’s development team and Musk began to tackle self-driving cars, the first place they went for advice and inspiration was Google. Google and its self-driving technology subsidiary Waymo have a solid history with self-driving cars. Google’s self-driving cars have clocked over 500,000 miles without an accident caused by a computer-controlled vehicle. (Source: Insider)
Even though Tesla is one of the most famous names in self-driving cars due to the viral popularity of the brand, it’s far from the only company to be pursuing self-driving car technology in the 21st century. According to market surveys, Tesla over-promises and under-delivers in self-driving capability compared to lesser-known AI companies.
Here are a few of the other major tech companies that have been challenging Tesla for the autopilot crown since 2013:
- Baidu: Baidu is the biggest autonomous technology company in China working with automobile makers such as BMW, Volkswagen, Ford, and several major Chinese car manufacturers. Baidu makes up Tesla’s major competition for self-driving vehicle markets in China and Southeast Asia.
- Argo AI: Argo AI is a self-driving artificial intelligence start-up backed by both Ford Motors and Volkswagen. While these two major automobile competitors make strange bedfellows, Argo AI has their confidence is a good sign for the company’s future.
- Waymo: Waymo is Google’s subsidiary company responsible for its autonomous technology division. The main difference between Google’s self-driving cars and Tesla’s is that Google offers self-driven car tours without a safety driver. Google is the only company that has managed to pull this off except in very limited cases.
- Nvidia: Many consumers may recognize the name Nvidia in connection with their graphics processing technology in personal computers systems, but Nvidia is also the chosen autonomous technology integrated by both Mercedes-Benz and Tesla itself.
While these companies may not be as well-known or have the marketing clout and hype that Tesla has, the AI technology being pushed by these corporations is pushing the envelope for what self-driving cars can do.
After public acknowledgment of the importance of self-driving autonomous vehicles moving forward, it wasn’t long before Musk introduced the first iteration of hardware into new Tesla models that would make it possible to later upgrade the vehicle with self-driving software. Below you’ll find a table of the technical specifications for Autopilot hardware 1.0.
|Digital Operating Platform||MobilEye EyeQ3|
|Forward Radar||525 feet|
|Forward Cameras||1 (monochrome)|
|Number of Sonar Sensors||12 (16-foot range)|
The first autopilot hardware suite designed for Tesla was designed by the Israeli engineering firm MobilEye. This company continued to work as Tesla’s partner in developing their self-driving technologies from 2014 through 2016, when they broke ties with Tesla over safety concerns. The Israeli firm felt that Tesla was moving too fast in progressing their self-driving technology.
While Tesla’s vehicles were equipped with the hardware to support autopilot and self-driving capabilities in 2014, the cars weren’t initially equipped with the software necessary. The hardware was just included as part of the overarching Tech Package to pave the way for semi-autonomous driving upgrades later.
It didn’t take long for it to happen, either. By the fall of 2015, Tesla was ready to drop their first major autopilot software upgrade to go along with the previous hardware upgrades included in the Tech Package: Tesla Software Version 7.0.
One of the most exciting things about the early years of Tesla’s Autopilot program was that the system could be released via remote satellite uplink to tens of thousands of vehicles at once. This system upgrade was pushed overnight to existing Tesla customers with vehicles with the hardware capability to support it.
However, some confusion about the nature of Tesla’s autopilot system led to mixed reactions at first. While the term “autopilot” tends to be associated with a system that can completely operate itself, Tesla cars are not safety-rated to be driven or operated without a driver behind the wheel and in full control of the vehicle at all times.
Autopilot did not (and does not) have the capability to drive a car without a manual operator. Drivers can remain much more relaxed in autopilot mode, but the car will still signal for the driver to touch the wheel if they let go of it for more than a few moments at a time.
The lack of hands-free driving technology isn’t a serious fault against Tesla, however. There are currently no self-driving car companies on the market that have reached a level of driving autonomy that removes the need for a human driver.
Even when self-driving cars do have the ability to drive with complete autonomy, it’s unlikely that government regulations worldwide will allow cars to forgo a person in the driver’s seat. For legal reasons, the chance for the driver to take a nap in the backseat while their car escorts them to their destination is a long way off yet.
Using the data gathered by beta testers throughout 2015, Tesla quickly got to work on their next iteration of their Autopilot program, Tesla Software Version 8.0. Here are a few of the upgrades that Tesla included in their newest version of the Autopilot system (Source: BGR):
- Expanded maps and navigation system
- Improved visual feedback for less distractions on the road
- Refined voice command system
- Cabin overheat protection for pets and kids
- 3D signal-processing
- Improved Autosteer response
- Fleet-learned curve speed adaptation
The technology included in the 8.0 update made the Tesla’s Autopilot features much more effective in low-visibility environments.
The major advantage that Tesla has in networking all of its fleet vehicles together on a single AI system. This networked system continuously learns from all the different Tesla cars out on the road each day, continually sending feedback for system refinements.
With a robust version of their Autopilot software system up and running, Tesla found itself needing to upgrade the included hardware in their vehicles to help support the new technology. In response to this, the company announced that all new and upcoming Tesla models would come with the hardware built-in to support full self-driving autonomy.
Below you’ll find a table of the technical specifications for Autopilot hardware 2.0.
|Digital Operating Platform||Nvidia Drive PX 2 AI|
|Forward Radar||525 feet|
|Forward Cameras||3 (narrow, main, wide)|
|Number of Sonar Sensors||12 (26-foot range)|
In addition to these improvements, the 2.0 hardware also got the following upgrades:
- Forward-looking side cameras
- Rear-looking side cameras
- Front/side camera color filter array
These additional cameras on new Tesla vehicles as well as their extended sonar sensor range and color filter array made the camera-based autopilot system much more accurate. In fact, the camera system in the Tesla Autopilot suite became so refined over time that the car dropped all radar sensors in 2021. (Source: Reuters)
In February 2017, Tesla pushed out new upgrades for its Autopilot program that converted it from Autopilot into Full Self-Driving (FSD) Mode. Despite the name, Tesla’s FSD mode is still only rated at Level 2 autonomy on the 1-5 scale designed by the Society of Automotive Engineers (SAE).
The newest version of Tesla’s self-driving technology includes the autosteering and traffic-aware cruise control of the earlier Autopilot version. It also includes many new features, such as the following upgrades:
- Summon/Smart Summon: Tesla’s Summon and Smart Summon capabilities allow a Tesla AI-enabled vehicle to back itself out of a parking spot unassisted and greet its owner at the curb. Smart Summon is an advanced upgrade of this mode that increases the car’s ability to recognize parking lot obstacles and manuever around other vehicles.
- Navigate: Navigational systems were already included in earlier versions of Tesla’s Autopilot, but the Navigate function in FSD is a much more comprehensive version of the software. Using this software, a Tesla vehicle can move itself on and off the freeway, gauge merges in traffic, and automatically engage the turn signal in autosteer.
- Automatic Lane Changes: Automatic lane changes is a complementary system to the Autosteer feature in Autopilot that allows Tesla vehicles to self-correct and maintain a lane in traffic or smoothly transition into a new lane without driver input. This system is dependent on Tesla’s camera and sensors system for knowing when to move the car.
- Traffic Sign Control: One of the biggest challenges of earlier Tesla Autopilot programs is that it had some difficulty recognizing traffic signals and signs on the road without a color filter array on its camera system. With upgraded cameras and software, Tesla’s FSD is much more capable of distinguishing between traffic signs and acting on them.
- City Streets Autosteer: The original version of Tesla’s Autosteer feature was reliable enough to be used on open freeways and interstates with driver guidance, but it didn’t operate as well at lower speeds and in more congested city traffic. The City Streets Autosteer update in FSD mode makes it easier for Teslas to self-drive in town.
- Autopark: Autopilot is a software feature that allows the Tesla to automatically parallel or perpendicular park. Once activated, this mode can park the car without any adjustments or inputs from the driver. Autopark is perfect for sketchy parking situations where human error can easily lead to a fender bender.
The FSD upgrade for Tesla’s Autopilot system costs an additional ten thousand dollars on top of the base price of the vehicle. For many consumers looking forward to Tesla’s fully-autonomous vehicles, the added cost is worth being one of the first to adopt this cutting-edge technology.
Other than an adjustment to the range of the Forward Radar and a secondary computing node added to the self-driving computer to increase computational efficiency, the hardware of Tesla’s FSD system didn’t change again in any serious way until another major upgrade was made in March 2019.
In this upgrade, Tesla implemented its own “system on a chip” that allows Teslas to link up with trained neural networks that have been refined by Tesla’s beta testing from 2018 through 2019. This neural network accelerator contains two AI chips for redundancy and is capable of processing data at a speed that is twenty-one times faster than the 2.0 hardware.
In September of 2020, Tesla started beta-testing of its new full self-driving software with the warning that the system was not yet capable of driving without driver input. Instead, the software was released onto the Tesla network so that it could be allowed to learn from the Tesla fleet to increase its efficiency as a system.
At first, Tesla only beta-tested its software with a few chosen beta drivers who were picked for their clean and careful driving records. However, as of July 2021, the newest version of Tesla’s FSD system is in full beta, meaning all Tesla owners with an FSD-enabled system are capable of enabling FSD features in beta mode.
From late 2020 through mid-2021, Tesla’s full self-driving autopilot has been under beta testing by the public. During this time Tesla has been under scrutiny about the safety of its autopilot testing on public roads without sufficient safeguards (at least according to some safety analysts).
So how safe is Tesla’s FSD technology? In the eight years since Autopilot technology was first designed and introduced, there have been only eleven recorded deaths in Autopilot-related incidents over nine wrecks.
Compare this to over 32,000 people killed in conventional automobile accidents each year (Source: CDC) and you’ll see that statistically, Tesla is still one of the safest cars on the road. While the idea of a car crashing on autopilot is frightening, even if you do crash in a self-driving Tesla, the safety features of the vehicle are likely to prevent any serious injury.
The biggest factor that impacts the safety of the FSD system in use is whether or not the driver is actively guiding the system. Some of Tesla’s most sensationalized crashes have been the result of drivers leaving the driver’s seat during Autopilot mode. (Source: CNN)
The reason that Tesla has been able to continually improve its artificial intelligence network over the past eight years is because of the fleet concept. Because each Tesla vehicle is connected to every other Tesla vehicle while driving, the system is constantly absorbing and integrating data from every driving session a Tesla engages in worldwide.
By activating Autopilot mode while driving the car normally, Tesla drivers are able to “train” their vehicle to learn roadway curvatures, adjust acceleration/deceleration, and refine lane corrections. The system is then able to project this information back to the fleet so that it can be integrated into all units.
The Autopilot/FSD mode doesn’t even have to be actively engaged and controlling the car to gather data for performance improvement. It can collect data as a backseat driver in a feature known as Shadow Mode.
The major advantage of Shadow Mode is that it allows Tesla’s networks to continue to gather data to improve their self-driving systems even if the owners don’t want to let the self-driving system take control of the car.
Here are a few more advantages to Tesla training its AI in Shadow Mode:
- Millions of miles of data: Through the computational processing power of its artificial intelligence network, Tesla is able to collect millions of miles of data every day as drivers use vehicles in the fleet.
- Active safety testing: Along with the information collected by the AI network during Shadow Mode, Shadow Mode allows Tesla to do massive amounts of safety testing on public roads. This allows the company to know that AI improvements have been tested in self-driving mode thousands of times before they’re ever implemented to the full fleet.
Tesla has been working on its autopilot program for almost a decade now, and the question has remained the same from enthusiasts to Tesla this entire time: When will Tesla self-driving cars be able to drive themselves?
According to Tesla, its fleet of self-driving cars will not be able to support a higher level of autonomy than Level 2 until the vehicles have experienced one to two million miles of aggregated data without driver intervention. (Source: Electrek)
Tesla’s capacity for self-driving autonomy in its vehicles has grown in leaps and bounds since Elon Musk first teased it in 2013. Tesla’s “full self-driving” cars don’t quite fit the bill for full autonomy yet, but it’s only a matter of time between Tesla losing the mandatory driver and self-driving cars taking over for good.