Statistics show that driving a Tesla with autopilot is as safe as it is without it.

image

Accident graph of driving with autopilot on and off in Tesla cars, adjusted to compare city and highway driving

Every quarter, Tesla publishes statistics on the mileage of the entire fleet, which falls on one accident. The reports separately show the mileage with the autopilot on, without the autopilot and without the autopilot and any other accident prevention systems. The latter is quite rare, so an interesting question arises: “Which driving is safer – with the autopilot on or off?”

About the author: Brad Templeton – Software Engineer, Robocars Evangelist since 2007, worked on GoogleCar in its early years. Founder ClariNet, honorary chairman Electronic Frontier Foundation and director Foresight Institute, founder of the faculty at Singularity University

Tesla’s reports are misleading to readers, suggesting that “driving with autopilot on is much safer.” That’s not true – according to the last quarterly report, driving with the autopilot off was slightly safer. Last quarter the numbers were roughly the same, in the first quarter of this year, driving with autopilot was slightly safer, and in all other quarters, driving with autopilot was slightly more dangerous. Of course, it should be borne in mind that there was a lockdown in the last three quarters, due to which traffic decreased and there were fewer accidents.

Here is a part of their report:

“In the third quarter, there was one accident for every 7.4 million kilometers traveled with the autopilot on. Drivers who drove without autopilot, but with safety systems on, had an accident every 3.6 million kilometers. Drivers who drove without autopilot or safety systems had an accident every 2.88 million miles. For comparison, according to reports from the National Highway Traffic Safety Administration, accidents occur every 771 thousand kilometers in the United States. “

Upon first reading, it might seem that the statistics of driving with autopilot are twice as good as statistics without it. The problem is that this is to be expected, because according to research from the Massachusetts Institute of Technology, 94% of the autopilot systems are used on highways with limited access (in fact, Tesla only allows the autopilot to be used on highways). Expected that new FSD system (a number of users are currently testing its beta tests) will be able to work in a similar way on city streets. Driving without autopilot consists of 40% highway driving and 60% city driving (the percentage for highway driving is lower as most drivers turn on autopilot). Most importantly, 1.6 km streets have 3 times more accidents than highways. Highway driving is much easier (so now the autopilot can only be used on them), and although accidents on them can be more serious, they happen less often (3: 1 ratio obtained from death statistics, I am still looking for more accurate data on all accidents) …

Thus, if you look at the statistics of the accident rate of the mode, which is used mainly on the highway, and the mode, which is used on the highway and in the city, the former will look better. With simple calculations, you can find out the real mileage for an accident – we get about half a kilometer.

In the Q3 2020 report, we will see that autopilot-enabled accidents occur every 8.28 million on the highway and every 2.7 million kilometers in the city. Combining these figures (in a ratio of 94 to 6) and we get the same “7.4 million kilometers per accident” from Tesla reports. If you turn off the autopilot, accidents will occur every 8.64 million kilometers (4% less often). This is just an estimate based on the available data. Tesla will have data on accidents on various types of roads, but the company has turned down several requests for their provision. 4% are indeed within the margin of error, so the numbers can be considered roughly equal. In fact, the error is so significant that other quarters may be equal to each other. However, the general trend is that driving with autopilot is slightly less safe than without it. Indeed, the dramatic jumps between the numbers published by Tesla suggest a lot of randomness in the results – the differences in quality are probably not as strong. In addition, we (to some extent) assume that autopilot systems have roughly the same ratio of accidents on highways and streets, but this does not greatly affect the overall picture, since autopilot is rarely used on city streets. Autopilot is probably a little more dangerous off the freeway and a little safer on the freeway.

If we manage to reduce the calculated accident rate per kilometer when driving on a highway without an autopilot to 2.3: 1 (the numbers are taken from other sources), then it will be possible to see on the graph that, until the first quarter of 2020, driving without an autopilot was slightly less safe, and then became more safe (by 15-20%) during the year. Perhaps it’s a lockdown or software development – both of them affect the results.

Looking at such a small difference, some may associate this problem with people who do not follow the road (although they must) when they have autopilot on. We’ve seen some of the drivers die or put themselves at significant risk when using autopilot, while diligent drivers can use autopilot and drive with the same (or better) level of safety.

There are a number of other oddities in the Tesla report. For example, it gives the number of “accidents” – as they call the activation of safety equipment (airbags or seat belts) in collisions at speeds over 19.3 kilometers per hour. According to other sources, airbags deploy approximately every 5.6 million kilometers for all cars – an unfortunate discrepancy. It is also reported that the statistics of “accidents” provided by the Highway Traffic Safety Administration differ significantly from those given in the Tesla reports. Management statistics are based on police reports. However, these numbers should not be compared with those of Tesla. Tesla recognizes this (in small print), although it should be more clearly stated.

Consumer Reports lowers Tesla score for lack of driver tracking systems

In recently published Consumer Reports rating put Tesla’s autopilot in second place among similar systems (first place went to Super Cruise from GM). The material claims that Tesla’s performance is better (although the margin is small). The authors also accuse the company of not having camera-based driver monitoring systems in their cars. Super Cruise has such a system, it makes sure that the driver’s eyes are directed to the road. Also, Tesla’s system works more explicitly and does not impose big restrictions (Autopilot can be turned on on any roads that seem normal, even including those roads that the system may not cope with. Super Cruise will only work on those roads for which there are special maps ).

These are very interesting estimates. Buyers rarely give a product higher ratings for limiting their ability or forcing them to use it correctly, but CR believes this is important in their ratings for improving overall safety. Tesla monitors the driver by tracking random steering wheel forces – they make the system know that the driver is not taking their hands off the wheel. Tesla does not track the direction of your gaze. New Tesla models have a camera in the cabin that can follow the driver. Tesla’s engineers have reportedly tried using these cameras to track driver’s condition, but this feature has not yet been announced. Some drivers do not want to be monitored for their safety and their driving, while others welcome it.

It’s worth noting that good driver tracking systems can improve Tesla’s safety record above. Often they are reduced due to accidents caused by drivers who do not follow the road properly. Even an eye tracking system could be useful (although Consumer Reports probably won’t approve of its optional inclusion). Tesla could well have made its autopilot only work on limited access highways. Based on past experience, Tesla is poised to challenge Consumer Reports. All other companies received fairly low marks, with only the Lincoln / Ford Co-Pilot scoring 50 points from CR (on a 100-point scale).


  • Russia’s first serial control system for a dual-fuel engine with functional separation of controllers
  • In a modern car, there are more lines of code than …
  • Free Online Courses in Automotive, Aerospace, Robotics and Engineering (50+)
  • McKinsey: rethinking electronics software and architecture in automotive

image

Vacancies

NPP ITELMA is always glad to young specialists, graduates of automotive, technical universities, as well as physics and mathematics departments of any other higher educational institutions.

You will have the opportunity to develop software of different levels, test, launch into production and see in action the finished automotive products that you had a hand in creating.

The company has organized a special testing center, which makes it possible to conduct research in the field of ICE control, including in a car. The testing laboratory includes motor boxes, drum stands, temperature and climate control units, vibration stand, salt fog chamber, X-ray unit and other specialized equipment.

If you are interested in trying your hand at solving the problems that we have, write in a personal.

About ITELMA

We are a large development company automotive components. The company employs about 2,500 employees, including 650 engineers.

We are, perhaps, the strongest center of competence in the development of automotive electronics in Russia. Now we are actively growing and have opened many vacancies (about 30, including in the regions), such as software engineer, design engineer, leading development engineer (DSP programmer), etc.

We have a lot of interesting tasks from car manufacturers and concerns that are driving the industry. If you want to grow as a specialist and learn from the best, we will be glad to see you in our team. We are also ready to share our expertise, the most important thing that happens in automotive. Ask us any questions, we will answer, we will discuss.

List of useful publications on Habré

  • Free Online Courses in Automotive, Aerospace, Robotics and Engineering (50+)
  • [Прогноз] Transport of the future (short-term, medium-term, long-term horizons)
  • The best materials on hacking cars from DEF CON 2018-2019
  • [Прогноз] Motornet – data exchange network for robotic transport
  • Companies spent $ 16 billion on self-driving cars to capture $ 8 trillion market
  • Cameras or lasers
  • Autonomous cars on open source
  • McKinsey: rethinking electronics software and architecture in automotive
  • Another war of operating systems is already underway under the hood of cars
  • Program code in the car
  • In a modern car, there are more lines of code than …

Similar Posts

Leave a Reply