[ad_1]
Waymo's self-driving cars traveled 10 million kilometers on public roads, the company said on Wednesday. This places the company very far ahead of its rivals. Only one other company has announced more than one million kilometers of tests. And it was Uber, which announced 2 million kilometers last year but was forced to freeze its public tests after a fatal accident in March.
This milestone is one of many signs that Waymo holds a significant lead in the race to market driverless technology. Waymo has announced plans to launch a taxi service in the Phoenix area before the end of the year. GM, one of Waymo's major rivals, aims to launch a commercial taxi service next year. Most other companies working on driverless technology do not plan commercial launches until the early 2020s.
Rivals say test miles are overvalued metric
If you talk to Waymo rivals, most of them are quick to deny the importance of testing miles. They argue that quality is far more important than the amount of driverless testing. According to them, it is easy to travel many miles on the same routes, but that may not prove much of the quality of its technology. The fatal accident of the Uber car is a typical example: at the time of the accident, Uber was only ahead of Waymo in terms of test miles, but his technology was deeply failing.
This argument was forcefully presented by Kyle Vogt, CEO of Cruise, GM's driverless car manufacturer, which is arguably the most formidable competitor of Waymo. Cruise's latest report to California regulators, filed earlier this year, indicated that the company had traveled only 131,000 km on public roads between December 2016 and November 2017 (for comparison, Waymo has accumulated 350,000 in California and many others in Arizona). But the bulk of these tests took place in the city of San Francisco and Vogt pointed out in a blog post last year that San Francisco is a much harder environment for driverless cars than the region from Phoenix, where Waymo does a lot of testing.
"Our vehicles face difficult (and often absurd) situations that can go as far as 46 times Vogt wrote: Vogt specifically compared the tests of Cruise in San Francisco and those in the suburbs of Phoenix, where Waymo (and at the time Uber) did much of their testing.
This argument has a lot of merit. One of the main goals of testing in the real world, especially early in the development process, is to deal with difficult and unusual situations. And a car will obviously encounter more tricky situations per kilometer in San Francisco than in Chandler, Arizona.
Earning miles is essential before a commercial launch
But the test goal changes a bit when a company is about to launch a commercial service. In the final phase trials, one goal is always to deal with difficult and unusual driving situations that can be incorporated into a simulator to ensure that all subsequent versions of the software can cope with them. However, the final phase tests also serve as a "dress rehearsal" for all aspects of the commercial system that will be launched soon.
End-of-phase testing helps to ensure that the sensors and other critical hardware of the vehicle have a low failure rate. It also ensures that remote fleet operators are well trained and that the organization is ready to face any eventuality. It allows the company to verify that cars will use safe locations to pick up and drop off passengers. And this allows the company to identify other non-safety related issues that may arise, such as the number of vehicles available for rush hour service and setting up plans. adequate for the cleaning and maintenance of vehicles.
This type of final phase testing should be as similar as possible to the final product delivered. It is also important to do this by transporting real passengers on "real" roads. This will inevitably involve repeating the same routes as current customers tend to travel to the same locations (work, school, local grocery store) again and again. This type of test will not provide as much useful data from the point of view of pure software engineering, as most kilometers will be on routes and situations that the software has already mastered.
But it is also difficult to think of an alternative. A driverless taxi service is a complex and integrated system, with hardware, software and Waymo staff, all working together to take passengers from one point to another. testers will not be able to discover problems resulting from the interactions of these systems – for example, Waymo employees fail to replace faulty components or make mistakes due to confusing backend software interfaces.
At the end of the day, there is no substitute for testing the system and seeing what kinds of problems come up.
People sometimes blame Waymo for focusing so heavily on testing in the Phoenix area, arguing that it would leave Waymo's cars unprepared for the harder driving environments elsewhere in the country. But that's exactly the point: Waymo's commercial taxi service will initially only be available in the Phoenix area.
No matter how the system would work in snow jams in Manhattan or Detroit, because, at least initially, Waymo will not offer commercial service anywhere other than in the Phoenix metro area. The company will likely do a lot more testing in those countries before they start services there.
In short, what strikes Waymo is not just the fact that he has done many more tests than any of his rivals, it is that the nature of his tests is different. Most of Waymo's competitors are still doing preliminary tests, the purpose of which is to build a library of unusual driving scenarios in order to integrate them into their simulators and to improve the software. Their tests are aimed at providing engineers with the data they need to improve the software.
Waymo, on the other hand, believes that his software is about to be ready for commercial launch, which has led the company to focus on more comprehensive tests, designed to validate the material, the organization and Waymo software. The fact that no one else is actually doing it seems to indicate how far Waymo is coming.
Source link
