In July of this year Cruise, General Motor Co’s self-driving unit, delayed the commercial deployment of cars past its target date of 2019 as the vehicles required more testing before they could be safely on the road. In a blog post on the news Dan Ammann, Cruise’s CEO, wrote, “When you’re working on the large scale deployment of mission critical safety systems, the mindset of ‘move fast and break things’ certainly doesn’t cut it.”
The topic has been back in the news recently, as the U.S. Senate Commerce Committee will hold a hearing on the numerous factors involved in deploying self-driving vehicles—including testing. The Chairman of the National Transportation Safety Board (NTSB) will participate, along with representatives from the National Highway Traffic Safety Administration and the U.S. Transportation Department. The former has been reviewing the fatal March 2018 crash of an Uber self-driving vehicle, and according to Reuters’ David Shepardson, “The NTSB may use the findings from [the] accident to make recommendations that could affect how the entire industry addresses self-driving software issues, or to regulators about how to better oversee the industry.”
To describe the technology environment of autonomous vehicles as complex is a wild understatement. To start with, the software behind self-driving cars must work reliably, integrate seamlessly with other technology, perform well in a variety of high-stakes environments and other requirements typical of mission-critical technology. And it only becomes more complex from there.
When on the road, self-driving cars must also account for hazards, weather conditions, other vehicles, and the personalities of the drivers behind them, to name just a few of the myriad safety considerations. The latter is something scientists at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) are attempting to solve by using machine learning to help cars predict motorists’ behavior and avoid those determined to be selfish or egotistical.
It’s an interesting concept, and certainly worth a read to learn more about the methodology and consider how the approach could be incorporated into autonomous vehicle testing and deployment. However, it also underscores just how much work remains until self-driving vehicles are ready for mainstream adoption—for example, the researchers behind the MIT CSAIL system acknowledge the technology is not yet mature enough to be implemented on real roads.
It will be interesting to keep an eye on this week’s Senate hearing and see what, if any, new legislation emerges around autonomous vehicles’ development and deployment. From our perspective as leaders in intelligent test automation, it seems pretty clear that testing must play a central role at every stage of the process.
Learn more about how Eggplant helps companies in the automotive sector modernize their testing approach, and you can also read about our work with mission-critical technology like NASA’s Orion spacecraft here.