After reading a variety of articles and discussion threads relating to the topic, I have concluded that GPS alone is nowhere near accurate, safe, or consistent enough to be used as a location sensor, or for the car to make safety critical decisions based upon.
I concluded this after reading here how the self-driving cars that Google has deployed across the US in recent years and reading how they have a variety of location sensing technology on board, including GPS alongside LiDAR, ultrasonic sensors, front and rear cameras for image recognition at very close distances, bumper mounted radar, gyroscopes, and tachymeters. To minimise the margin of error in the GPS data, the self driving car compares GPS data with data collected by other cars with specialist sensors in the same location, and checked for consistency before being used.
Additionally, I read a particularly interesting discussion thread on internet message board Hackers News which talked about GPS’ susceptibility to ‘jamming’ – whereby GPS signal could be jammed maliciously to prevent self-driving cars from navigating.
Because of the aforementioned limits with GPS, a British company called Oxbotica – a spin out of Oxford University – has developed software that allows self driving cars to navigate solely with LiDAR scans – I read about this here and found it to be a particularly fascinating solution.
To conclude, I feel that whilst GPS is not a satisfactory by itself; it is evident from my reading on the success of Google’s Self Driving car project ,that when coupled with data from LiDAR, radar, and accelerometer data, it becomes suitably fault tolerant and is acceptable for use in self-driving cars.