An recent article in the New York Times makes you sit up and take notice: Contrary to general expectations, self-driving cars would lie further in the future than had been assumed. The newspaper quotes managers of the automotive industry and AI developers who do not believe in an imminent revolution on the road. Isn’t the autonomous vehicle ready for the mass market yet?
The story is a bit reminiscent of the smart refrigerator, which is supposed to provide for supplies automatically. The first models have been making headlines since the end of the 1990s. But even 20 years later they have by no means arrived on the mass market.
Has the car industry overestimated itself?
In March 2018, Apple engineer Walter Huang died in an accident in California in his 2017 Tesla Model X when the vehicle crashed into a concrete crash barrier in autopilot mode without braking at around 109 kilometers per hour. Huang had switched on the autopilot 10 seconds before the fatal crash and then immediately took his hands off the steering wheel. In May of the same year, a Model X driver crashed, this time in Florida, while driving into a parked police car. Luckily, this time the damage to the sheet metal remained.
A year later, in March 2019, Jemery Beren Banner died in Palm Beach in his Tesla Model 3, when the vehicle also raced into a preceding truck 10 seconds after the autopilot was activated. The first victim of a Tesla in semi-autonomous mode was Joshua Brown. In 2016, his vehicle in Florida also collided with a truck crossing the road.
Dispute over the optimal recognition system
After each individual accident, the dispute breaks out anew: industry experts criticized the recognition system in Tesla vehicles. Tesla defenders, in turn, argue that Tesla has never claimed its vehicles to be suitable for fully autonomous driving. The driver must always have his hands on the steering wheel in order to be able to intervene in time. The drivers negligently overestimated the car and its capabilities.
A technical point of criticism at Tesla is the missing LIDAR recognition system in Tesla vehicles. The “Light Detection and Ranging” system is a radar-like system designed to help autonomous driving cars assess their surroundings and the behaviour of other road users correctly. Lidar technology can use laser pulses to scan its surroundings 300 metres away. This results in a three-dimensional image of the environment the vehicle can orient itself to. In addition to cameras, ultrasound and radar sensors, most vehicle manufacturers also use the LIDAR system for environmental detection. Only Tesla has done without it so far and has installed “only” 8 cameras, 12 ultrasonic sensors and a forward facing radar in its vehicles.
Incredible amounts of data, decisions in milliseconds
In order to capture and evaluate the incomprehensible amounts of data and then let the vehicle make decisions in fractions of a second, the computers installed in the cars must meet the highest standards. A cloud solution in which the data is first sent to a data center for evaluation is out of the question anyway, given the short response times required.
And here we are obviously faced with a bigger problem than we thought. Ford and other companies say that the industry overestimated the arrival of self-driving cars. The main problem was that it was still difficult to predict what other road users would do.
Self-driving buses are less of a problem here. They drive on fixed routes and at low speeds. It is much more difficult to give fast cars a safe behaviour of their own.
Now Ford and VW have teamed up to push the topic forward with the Argo AI startup from Pittsburgh, which specializes in autonomous vehicle technology. As early as 2021, cars were to be used autonomously for carpooling in some urban areas. Can the time frame be met?
The human factor
Argos boss Bryan Salesky describes the main problem of autonomous driving: the human factor. In road traffic, the participants did not always behave as desired. Cyclists on the wrong side of the road, cleaning vehicles unexpectedly changing direction. “With radar and high-resolution cameras and all the processing power we have, we can detect and identify the objects on a road. The difficult thing is to anticipate what they will do next,” he told the New York Times.
According to Salesky, about 80 percent of the technology needed to put self-driving cars into routine operation has been developed to date. But the remaining 20 percent, including the necessary software that reliably predicts what other road users will do, will become much more difficult.
Problem of micromaneuvers
“We overestimated the arrival of autonomous vehicles,” Ford’s CEO Jim Hackett said at the Detroit Economic Club in April.
The so-called “micromaneuvers” are also problematic, for example when vehicles drive more slowly in search of a parking space and the autonomous vehicle behind them has to follow more carefully and at a greater distance. Of course, any car can be programmed in such a way that in dangerous situations it always immediately steps on the brake. But an over-cautious car is also unacceptable for flowing traffic.
Maybe we need to lower our expectations a little. Self-Driving vehicles will come. Maybe it’ll just take a little longer.