Self-driving cars are no longer a thing of sci-fi movies or the future. Tesla Motors has some of the safest vehicles on the planet, partly because of the Tesla auto-pilot feature. Sitting back and letting your car do the driving after a long day of work sounds too good to be true. However, some people question whether an autopilot computer system is safer compared to a human being.
Tesla autopilot vehicles are some of the safest cars globally and have the lowest probability of injury or crash ever tested by government institutions.
Humans get distracted and make mistakes. They are also a single person with a single brain focused on the act of driving a car. However, the Tesla autopilot is always focused and uses driving information from millions of drivers across the globe that influences its decision-making on the road. Read on to find out more about Tesla autopilot safety compared to human drivers.
There is no question that the Tesla fleet of vehicles is full of incredibly safe cars. CEO Elon Musk says that the vehicle is around ten times safer than human beings driving cars. However, the numbers are slightly skewed as most vehicles in the United States are over twelve years old, making them more dangerous. Also, most autopilot driving gets done on highways and freeways, which offer safer driving conditions. Experts have pointed out other flaws in the system inherent in the unpredictability of driving in the real world.
In 2021, Tesla autopilot had only one accident for every 4.19 million miles driven. Compared to all injury crashes in the United States for all vehicles, Tesla has remarkably safe statistics. Take the numbers of accidents for all cars by age group listed below per 100 million miles driven:
- 16-17 age group: 1,432 crashes per 100 million miles
- 18-19 age group: 730 crashes per 100 million miles
- 20-29 age group: 549 crashes per 100 million miles
- 80-plus age group: 432 crashes per 100 million miles
The accident numbers from Tesla Autopilot extrapolated out from 4.19 million miles driven to 100 million miles driven equals about 24 crashes per 100 million miles driven. Eighty-plus-year-olds make up the safest age group with the least amount of crashes in the United States. Yet, they have a crash rate that is almost twenty times more than the crash rate of the Tesla Autopilot.
As you can see, the Tesla autopilot is incredibly safe when compared to all other age groups and all crash statistics of all automobiles in the United States. The small number of crashes by Tesla Autopilot is an excellent jump in safety from just one year ago. For example, in October 2020, Tesla Autopilot driving safety data showed that Autopilot was just as safe as human drivers. Now, just one year later, it is much safer.
Why The Tesla Autopilot is Safe
According to the numbers reported from government agencies and other organizations, such as AAA, the Tesla autopilot system is incredibly safe. The number of crashes, especially injury crashes, has decreased substantially in the last year. It seems like the Tesla autopilot is getting safer all the time.
Society of Automotive Engineers Defines Self-driving Tech
The definition of self-driving gets rated on a scale of 0 to 5. The Society of Automotive Engineers defines this technology as starting at level 3, and it gets labeled as “semi-autonomous driving.”
Overall, level 3 means that the car manages most parts of driving and monitors obstacles and the surrounding area. Some of the factors that go into level 3 autonomous driving include:
- Traffic chauffeuring
- Required conditions for driving must get met
- When the system requires, you must drive
- You are not driving when the features are engaged, even though you are seated in the driver’s seat.
You may have Tesla’s autopilot request for the driver to make some steering wheel maneuvers, so the driver must remain aware. Some have tricked the system by hanging weights from the steering wheel. However, driving in your Tesla without sitting in the driver’s seat is definitely not recommended. Suppose the driver takes their hands off of the steering wheel for too long. In that case, the autopilot system shuts off; in other words, the technology is not full-autonomous yet.
Several factors make the Tesla autopilot extremely safe. The most crucial safety factors include the constant real-time updates on the Neural Network, driving data of all other Tesla drivers, camera, and objection detection radar technology.
Computer learning based on data is one of the most vital attributes of the Tesla AI autopilot system. The general rule is that the more information that an AI system has, the better they function. Having said that, there are three significant parts of the neural network that all Tesla vehicles are constantly using for learning.
Some updates get pushed into the computers on-board Tesla automobiles. This data then gets used as part of the Neural Network. This network of learning and updating gets based on statistics from three main categories:
- Computer Vision
- Prediction Capability
- Path Planning or driving policy
The Neural Network is the interconnectivity of all Tesla vehicles. Each vehicle records interactions with unique driving conditions. These snapshots and recordings get uploaded to the Neural Network as a collection of data about:
- Driving conditions
- Driving behaviors
- Driving strategies
All of this data gets processed as a cohesive whole, almost like a singular “Tesla Brain” known as the Neural Network. However, this network gets made up of several different parts, all of which do various jobs to support the autonomous driving system and ensure that the system is constantly learning and updating.
The vision of the computer is just like the vision of a human driver. One of the essential parts of computer vision is detecting objects and obstacles in the pathway of the automobile. The goal of the Neural Network and computer vision is to understand the world around the car. Learning the world around the vehicle requires the system to take snapshots of objects in the real world so that they get stored on the network and analyzed for computer learning.
These snapshot of unknown or rarely encountered obstacles and objects is a constant task for the computer vision on Tesla. For example, if the Tesla faces something seldom seen, like a moose, it takes a snapshot and learns the various moose shapes from all encounters.
These snapshots are uploaded into the neural network many times from different encounters until the system learns what a moose is and a photo is no longer needed. The problem with this system includes the fact that objects are difficult to recognize by the computer. Also, labeling these objects is only possible on a large scale by humans. Finally, determining when an image is no longer needed is a long and arduous process that requires computer learning run by human engineers.
It is helpful that Tesla has drivers going billions of miles a year so that even rare objects, like moose, have snapshots taken of them. These rare objects are eventually learned by the neural network and help drivers of Tesla autopilot stay safe while driving.
The next level of complexity for the neural network is the capability for the computer to make predictions about interactions on the road. These predictions get based on situations already encountered by other Tesla drivers and recorded by individual Tesla’s then uploaded and learned by the neural network. Basically, the network is learning future predictions.
The behavior of cars, pedestrians, and cyclists on the road is very difficult for the AI to predict. But, after years of training with the recorded interactions of billions of miles driven, the network is getting a bit better.
Incorrect predictions, however, are also valuable learning information. When the car makes a mistake, the mistake gets registered with a snapshot and then uploaded to Tesla training sets. These training sets then get tested in the real world and on Tesla testing driving situations before being pushed out to the whole neural network.
The Neural Network and the AI software that runs the data processing and Tesla system updates are different than the computer learning snapshots, which mostly require human labeling. The neural networks are genius because the AI software can compile and learn the encounters and situations for prediction on their own. Behavior patterns or determining which behaviors precede other behaviors are part of the system for learning both common and rare occurrences.
The actual driving policy of the car is the actions that the vehicle is making while on Autopilot. Driving policy for the cars directions, actions, and roadway decisions might include:
- Driving the speed limit
- Staying in the center of the lane
- Passing other vehicles driving slowly
- Changing lanes
- Stopping for objects or jaywalkers
Driving policy is the most challenging learning that the vehicles’ Autopilot must undergo because of the nearly limitless possibilities for interactions on the roadways. One way that the Tesla neural network gets around this difficulty is by using imitation learning, which is basically where the systems of the car copy what humans who are driving Teslas on the road around them do.
There are many Tesla drivers in the world, and they are all connected via the neural network. In other words, all Tesla cars are taking snapshots, learning for roadway interactions, and then uploading and updating the Autopilot as a cohesive unit.
The driving data for Tesla is incredible and equals nearly 500,000 new vehicles on the roads every year. This large fleet of millions of Tesla vehicles compiles data on billions of miles of driving. The large volume of connected vehicles makes the Tesla AI system and neural network stronger than any other car company.
Driving data from other Tesla owners also gives your driving an edge. Compared to a single human driver, there is a wealth of experience and learning that the Tesla autopilot has at its disposal, which a human driver does not. The many different rare occurrences that a human driver may only face once in their life are experienced by the Tesla neural network many times.
In the end, analysis of driving data from other Tesla drivers makes driving in autopilot mode extremely safe. The added benefit of the computer learning based on data from every other Tesla driver gives you the confidence that the Autopilot knows what to do when encountering even the rarest of road obstacles and situations.
The cameras used for collecting data on the road and snapshots of new, rare, or commonplace events, situations, and objects are state-of-the-art. On the latest version of all Tesla’s, eight cameras get mounted around the vehicle in strategic locations. These locations are all ver the exterior of the car and include:
- Rearward-looking side cameras: These cameras are large and pick up obstacles and other cars in the lane to the side or behind you at a distance of up to 100 meters.
- Rearview camera: This is a slightly smaller and more detailed camera that works to detect imminent collision objects and aids in reversing the car and viewing things about 50 meters from the vehicle.
- Forward-looking side cameras: These cameras are incredibly wide and have an angle of more than 180 degrees of view. The wide forward-facing cameras see things about 80 meters from the car. They are mainly for lane changes, parking, imminent collision, and pedestrian or other road condition obstacle detection.
- Wide forward camera: This camera is mainly for doubling up the car’s front corners with sensors for imminent collision, parking, and moving obstacles. They have a range of about 50 meters ahead and to the car’s sides at a 45-degree angle.
- Primary forward camera: This camera drives and senses the main roadway conditions ahead of the car. It is also closely linked to the car’s steering, acceleration, and braking and has a sight range of over 150 meters.
- Narrow forward camera: The scouting camera of the car looks out at a distance of over 250 meters and signals possible obstacles ahead in the road. It has a narrow beam, so it works in conjunction with the other forward-facing cameras to detect what is coming directly at the front of the car as you are driving on the roadways.
Radar is another detection system that is part of every Tesla.
Tesla has recently dropped the use of radar and transitioned to only using the camera system, called Tesla Vision, for Autopilot. Since cameras are now the primary way that Tesla detects obstacles in the road during Autopilot, there is no longer a radar technology on Teslas.
The idea is that the Tesla Vision is better at adapting and using snapshots of road conditions and situations for computer learning. This enhanced camera use will bolster the neural network by focusing primarily on what the cameras see as the data that gets used in computer learning.
It gets reported that the camera-only system will limit the incidence of “phantom braking,” where Tesla vehicles suddenly brake while on the freeway and passing under an overpass. The camera-only system or Tesla Vision system limits the data noise that the system picks up and makes for a more smooth and enjoyable ride while in Autopilot.
Tesla Vision may replace the radar system. However, the radar used in Tesla was some of the first technology used since the inception of self-driving technology. Radar has been an integral part of the Tesla system since 2014, and even though it is going by the wayside, it has helped Tesla develop one of the most sophisticated Autopilots driving strategies in even known.
Tesla has gone through several alterations, price changes, and evolution of features for their autopilot technology. Over the years, these changes have led Tesla drivers to decide between affording the autopilot system or settling for driving without it.
The history of the system’s evolution starts from the early days of the first Tesla vehicles and runs up to the present. Also, the cost of the system has changed drastically over the years, from high-priced additions that cost thousands of dollars up-front to the new monthly subscription service for much less money.
The history of the costs of Tesla’s Autopilot has produced some different models of the same system. There was never an autopilot system or any software for self-driving included in any Tesla models before September 2014. The lack of autopilot versions consists of the first two models created by Tesla; the Model S and Roadster. However, since then, there have been several different renditions of the autopilot system with various costs associated with them.
The first version of Autopilot had some simple but powerful features. It got priced at $2,500 before delivery and $3,000 after delivery of your Tesla. Autopilot #1 was initially available from October of 2014 through October of 2016. It had only a few features that came with it, including:
- Adaptive Cruise Control
- Auto Steering
- Auto Park
The 1.0 version of Tesla Autopilot strangely only used the one front-facing camera along with radar. The use of only one camera was probably due to the limited data uploading in the earlier version.
The additions for the second version of Tesla’s Autopilot were the most significant jump seen to date. Not only was the driving more precise and safer, but moved from one camera to all eight getting used. The upgraded features available for Tesla autopilot 2.0 include:
- Adaptive Cruise Control
- Auto Steering
- Auto Park
- Auto Lane Change
- Auto Navigate
The addition of several hardware upgrades allowed for more computing speed in the car, faster promotions, and better use of all eight cameras for Autopilot. This increase in computing rate made it possible for all eight cameras to get used as part of the Autopilot system. This version was available from October 2016 alongside the full self-driving version, which came out simultaneously. This system was $5,000 before delivery and $7,000 after.
The third version of Tesla’s Autopilot is known as the first one that includes an entire self-driving system instead of simply offering driving assistance. Full self-driving came out at the same time as the enhanced Autopilot in October 2016. The design provided many of the features that Tesla Autopilot is now famous for, including:
- Adaptive Cruise Control
- Auto Steering
- Auto Park
- Auto Lane Change
- Auto Navigate
- Stop sign and traffic light control
- Steering on city streets
During this version of complete self-driving Autopilot, you could buy it before delivery for $3,000 or after delivery for $5,000. There was also a momentary month subscription of $99 that did not last long.
The current version of self-driving required an autopilot upgrade on your Tesla and was available starting February 2019. This version has all of the same features as full self-driving version 1.0, with the added computer learning of several years, of course.
The cost of this package includes Autopilot and is $10,000 before or after delivery. The monthly subscription is also available again at $199 per month.
The evolution of the Tesla Autopilot system has seen both hardware and software upgrades. The price and how you can buy these systems have also changed over the years. For a long time, a single price for the Autopilot system was all you could purchase. It usually costs a little more if you ordered the system after the Tesla had already been delivered to you.
In October 2016, a monthly subscription was started for a short time then disbanded. However, beginning in 2019, the monthly subscription returned. The current monthly subscription gives you access to the Autopilot driving system and the self-driving system for one monthly price of $199.
The monthly subscription is uploaded into your computer system in your own personal Tesla vehicle and can get canceled at any time. Paying for the technology by the month may make more sense since you have no guarantee that paying full price for the Autopilot technology will increase resale or even get included in a used car.
The question of who is at fault when the Tesla autopilot crashes is a philosophical question that brings up some serious dilemmas. The company always tells drivers they need to be in the driver’s seat, holding the steering wheel and paying attention when the Autopilot is engaged. However, losing focus while the car is driving for you is nothing new and not surprising.
Usually, the human being is at fault when there is a fatality involving autonomous driving. However, Tesla may push the line a little bit by allowing drivers time for daydreaming or letting go of the wheel. Compare that to the BMW or GM semi-autonomous driving systems, which have cameras in the steering wheel which make sure you never take your eyes off the road.
Overall, some tests have shown that Tesla cars get easily tricked into driving automatically without hands on the wheel or even someone in the driver’s seat. The system can get fooled relatively quickly and points towards Tesla shouldering more responsibility for crashes than previously recognized.
Elon Musk, the CEO of Tesla, once famously predicted that a cross-country trip would be possible without touching the steering wheel by the year 2017. The autopilot technology is far from this prediction, and most of the lag is due to safety concerns.
However, the hardware needed for full self-driving service is already present in most newer Tesla’s. What is required is better upgrades to the computer learning and processing of the data in the Neural Network.
The promise of Autopilot safety from Tesla is real and will happen one day; the only question is when. The amount of driving data accumulated by this point by Tesla is immense. It shows that the fully autonomous driving car may come sooner than later. But, the self-driving vehicle may never be 100% independent.
The “edge” or “corner” or “fringe” situations that require split-second human calculations and reactions are not programmable and can’t get handled by a computer. So, if the promise is for an almost entirely automatic autopilot, the guarantee of Tesla Autopilot safety is excellent.
Right now, Tesla vehicles drive more safely than human beings. However, whether or not the cars that Tesla produces will ever achieve complete self-driving technology is up for debate. What is clear is that the addition of cameras, radar, and a worldwide neural network of computer learning and situational driving data makes Tesla cars incredible safe.
Human beings are always prone to disastrous and tragic miscalculations, sometimes at the most inopportune time. However, humans also have lightning-fast reactions that save lives on the road. Computers are steady machines that can get programmed very well for all general road conditions. Still, they may not react well to unknown situations. Overall, Tesla offers some very safe autopilot technology for its drivers.