Generally, self-driving cars can be safer than human drivers. Unlike humans, many have 360-degree sensors that are active constantly, and can be effective at eliminating most accidents caused by human error. Automated Driving Assist Systems (ADAS) systems never get tired, do not text while driving, do not drive under the influence, and can reduce the number of crashes caused by human error.
However, driverless cars can still trip up under certain situations. Bad weather can impair the car’s sensors and systems, making them more prone to errors. Driverless cars are still learning to navigate the chaos of inner-city traffic, especially in large megacities like New York, Boston, and others. Additionally, driverless cars perform worse in edge cases (extreme or highly improbable situations that are not normally expected in the real world).
Most safety regulations for automated vehicles are not mandatory, and regulators are unsure of how to standardize laws across states. Computers lack intuition and instinct which is common in human beings. This can affect their response in some situations. Thus, self-driving cars require further engineering to improve their response to complex situations – particularly edge cases. In many cases the sensor systems can perceive and act faster than humans in certain operation design domains. We may still need many years of improvements before they can be truly termed ‘safer’ than humans in the majority of driving conditions.
Driverless taxis and trucks have logged millions of miles on the road with and without drivers, with billions of miles in verified simulations. When operated appropriately in accordance with the vehicle’s operational design domain (ODD), they have fewer crashes than humans. Results from the first set of data from NHTSA’s standing general order show that the vast majority of crashes involving an ADS-equipped vehicle were from the vehicle being hit from the rear.
Summary Report: Standing General Order on Crash Reporting for Automated Driving Systems.