That’s a great question.

In this case, it’s relatively easy to simulate sensor malfunction (and make it happen more frequently in simulation). On the other hand, the most sensitive system to cosmic rays would be the large system(s)-on-chip(s). Luckily, in the example given above, the simulations are running on real hardware, subject to real cosmic rays. Assuming that simulation is running in real-time, which you would expect to be true if the compute system is sized to match the real-life workload, then the rate of errors from cosmic rays (which is very small) would be the same as in real autonomous vehicles.

Accounting for factors like cosmic rays is something that I used to do when I designed on-silicon logic, and it’s something that I believe is accounted for in the safety equation for automotive systems. However, I think that when it comes to autonomous vehicles, there are much more commonly troublesome factors that need to be accounted for in simulation to increase robustness in real-life, such as getting mud and water spatters on camera lenses.