Once again, many thanks to Tom Simpson in our autotech team for this piece.

In March, an Uber self-driving car was involved in a high speed crash in Arizona in an event that highlights the challenges human drivers can pose to autonomous vehicles.

The Uber vehicle (which was in self-driving mode at the time) was cleared of fault by the police report.  The driver who caused the crash was trying to cross three lanes of traffic, and although the first two lanes were free, the driver failed to see the oncoming Uber vehicle in the third.  The impact resulted in the Uber vehicle crashing into a traffic pole and then spectacularly flipping on its side before colliding with two other vehicles.

Uber initially suspended its self-driving car operations while it investigated.  Three days later it cleared its autonomous cars to resume driving in three cities where it operates its self-driving pilot programme: Tempe, San Francisco and Pittsburgh. 

This accident is a reminder of one of the inherent challenges that driverless cars currently face: human drivers of conventional cars.  Human drivers do not always adhere to the letter of the law.  They may speed up to catch a green light or, as in this case, try to ambitiously cut across lanes.  How can engineers predict and counteract such behaviour?

Analysing the internal metrics of driverless cars after accidents like this one enables the overall safety of these vehicles to improve.  According to Hod Lipson, professor of mechanical engineering and roboticist at Columbia University, "driverless cars keep getting better the more they drive, whereas humans have a roughly constant safety record over the years".  In Lipson's view, the idea that "somehow a human driver makes the drive more secure is false comfort, and potentially dangerously misleading".

Do we subject driverless cars to higher safety standards than those driven by humans?  Probably, and rightly so - but we need to accept that their safety records can't remain unblemished, even though overall safety standards should improve markedly.