Three of the world’s leading automakers are joining forces to develop safety standards for self-driving cars. The new consortium will be called the Automated Vehicle Safety Consortium (AVSC), and it will be affiliated with the prominent auto engineering group SAE International.
The founding members are Ford, GM, and Toyota. All three companies have invested heavily in self-driving technology, making self-driving safety much more than just a theoretical concern for them.
GM is the parent company of Cruise, a self-driving taxi startup that is currently testing dozens of autonomous vehicles in San Francisco. Ford has a similar effort called Argo that’s testing in Miami and working to expand to Washington DC.
Toyota has been funding several self-driving efforts in parallel. Toyota has lavishly funded its Toyota Research Institute—a research center that is working on both crash-prevention systems for conventional cars and fully autonomous vehicles. Toyota also invested $500 million in Uber’s self-driving project last August.
As is often the case with this kind of announcement, it’s a little hazy what the new group will actually do. The group’s first product will be a “roadmap of priorities” that will guide subsequent efforts. In its introductory press release, the group says that it will develop a “safety framework” for autonomous vehicles.
Others are working to develop safety frameworks of their own. Intel’s Mobileye announced a framework last year that provided a detailed taxonomy of the situations a driverless car needs to be able to handle. When Nvidia announced a safety framework of its own last month, Mobileye blasted it as an inferior copycat. The new group might have had this kind of dispute in mind when it announced plans to “harmonize with efforts of other consortia and standards bodies throughout the world.”
The value of sharing sensor data
The AVSC’s press release did mention one specific goal that seems sorely needed: better sharing of sensor data. The group’s website says it will work on new methods for “data collection, protection, and sharing required to reconstruct certain events.”
Major self-driving companies like GM’s Cruise currently have dozens of cars on public streets to test the performance of their technology in real-world situations. This process is not only expensive in terms of money, it also inherently poses a risk to the public, as we saw in last year’s deadly Uber crash.
There’s probably no way to eliminate this risk entirely, but a good way to minimize it would be for self-driving companies to share data with one another. If one company’s car encounters an unusually tricky situation, it should be able to share detailed sensor data with other companies. Other companies, in turn, should be able to recreate the scenario in their simulators and verify that their own software would have handled the situation correctly.
This kind of sharing ought to accelerate the development of fully driverless cars and potentially save lives in the process. But so far the intense secrecy that traditionally surrounds technology projects seems to have hampered any serious efforts to do this kind of data sharing on a significant scale.
So this seems like an area where the new AVSC should be able to provide a tangible contribution that goes beyond roadmaps and vague safety guidelines. The group could build the standards and infrastructure for Ford, GM, and Toyota to share sensor data among themselves. The three companies could build up a library of standard simulated driving scenarios and invite other companies to join in.
