3 ms·
They need to do a complete verification of releases on core requirements. My understanding is that the vehicle that killed the pedestrian was configured in a w
by ilaksh 7y ago
They need to do a complete verification of releases on core requirements.
My understanding is that the vehicle that killed the pedestrian was configured in a way that allowed it to fail in a very basic way.
What they need is a way to ensure that the actual vehicle configuration is translated fully to the simulator, and to run through a set of basic scenarios with every release configuration. The equivalent of end-to-end tests that run as part of a continuous integration process.
I assume they have something like that, but some configuration aspect is not properly integrated into the simulation tests, or they sometimes skip the full test suite. Or somehow their test suite did not include jaywalkers.
That is what needs to be in place, and there should be employees of an actual federal agency verifying that process, not people who are in any way paid by Uber.
That actually should be applied to all self-driving companies. And down the line, there should be government-run simulations that verify releases pass basic scenarios. Which is going to be very difficult to build, but the government can get funding through taxes or something from the self-driving car companies. They could even mandate that Google and Tesla provide their simulation software to the federal government, and create interfaces that are compatible, and make every release run through both simulators.