Having lowered their expectations, Uber plans to slowly resume testing its self-driving cars on Pittsburg roads, eight months after a fatal accident in Arizona that saw the company and its program come under intense scrutiny.
In March this year, a Volvo SUV equipped with Uber’s autonomous-vehicle (AV) system failed to respond to a pedestrian crossing the road, striking and killing 49-year-old Elaine Herzberg at approximately 60 kph (see related article).
The crash caused the safety of autonomous vehicles, and specifically the LIDAR technology that drives them, to come under intense public scrutiny, with some suggesting Uber’s self-driving program was too aggressive.
Now, a downgraded version of the original program will begin in the next few weeks, and will see Uber’s autonomous vehicles run a one-mile loop in Pittsburgh. The cars won’t operate at night or in wet weather, and won’t exceed 25 miles an hour (40 kph).
Before the accident, Uber had around 200 autonomous vehicles on the roads in Arizona, Pittsburgh, San Francisco and Toronto, with vehicles clocking speeds as high as 55 miles an hour (89 kph).
The program has continued to face obstacles since its inception, with cars failing to react as fast as human drivers, and some test drivers complaining that software patches were causing the cars to drive erratically.
An internal email sent to Uber executives that was recently made public raised safety concerns about the company’s autonomous vehicle program just days before the fatal accident in March.
The email claims that Uber’s vehicles were getting into accidents regularly, with the fleet “hitting things every 15,000 miles”, and vehicles being damaged “nearly every other day”. It also claims that near misses occurred every 100 miles and that backup drivers had to intervene once every one to three miles.
Last month, Uber chief executive Dara Khosrowshahi noted that the company recognises that there are still considerable inherit risks involved in their self-driving program.
“We are committed to anticipating and managing risks that may come with this type of testing, but we cannot — as no self-driving developer can — anticipate and eliminate every one.”