The facts date back to June 3. As every day since February 2022, one of the Cruise company’s electric Chevrolet Bolts has been criss-crossing the roads of the city of San Francisco in California. But while negotiating a left-hand bend, the self-driving car was hit by an oncoming Toyota Prius in the opposite lane. Fortunately, there were no serious injuries, but the responsibility of the autonomous car was immediately questioned. Very quickly, the main fault fell on the driver of the other vehicle, a Toyota Prius driving above the speed limits (40 mph in a zone limited to 25 mph, i.e. 64 km/h instead of 40 km /h). Especially since he was in a lane reserved for vehicles wanting to turn right and he continued straight, thus causing the accident with the autonomous car which blocked his way.
But the reactions of the robotaxi operated by the Cruise company raise questions. The autonomous car detected the Toyota Prius arriving in front. As the latter was driving in the right lane reserved for vehicles to turn, its software interpreted that the vehicle was going to turn right just behind it. As it approached its left turn, the self-driving car then detected that the Prius was continuing straight ahead contrary to what was expected. She then braked hard without this avoiding the accident, hit from behind by the Japanese compact hybrid. And this is precisely the problem in this situation: in their investigation into the accident, the experts of the NHTSA (the American agency in charge of road safety) estimated that the software of the autonomous car did not succeed to see it coming. The self-driving car’s ability to cope with a sudden change in direction has been faulted. Even though this late change of direction is against the rules of the road, the accident could have been avoided with a better interpretation of the data. Would a real driver have succeeded in preventing the accident when he saw the motorist coming in his Prius? It is very possible.
A reminder to modify the software
Faced with the conclusions of the investigation, the company Cruise recalls that its autonomous car was not at fault and that it was the behavior of the other driver that caused the accident. But the company will be recalling all of its robotaxis in San Francisco to update their software to better respond to such situations in the future. Will artificial intelligence succeed in understanding and anticipating human unpredictability?