The world’s unpredictable
Maarten Sierhuis, director of the Nissan Research Centre, said: ‘Reasoning and learning is not simple. From the raw data, we need to perceive the world and understand the obstacles. It’s hard [for a machine] to distinguish a human standing still from a lamppost. Then the system needs to predict what the object will do next and what every other object will do.’
Real-world driving involves quick decision making that would require some extremely sophisticated programming. For instance, going through a red light because you’ve been waved through by a roadworker or a policeman. ‘When you drive, how many times do you break the law in traic? More often than you think.’
Drawing on his experience working with NASA, he believes that you need to keep humans in the loop, supervised by something like air traic control.