Unpredictable Londoners Are Training Driverless Cars to Deal with Human Behaviour

By Holly Brockwell on at

Londoners have always thought of themselves as a bit special, and a software startup at Imperial College has pretty much confirmed they're right.

Humanising Autonomy, based at the Imperial Enterprise Lab for supporting 'innovators and entrepreneurs,' is using London pedestrians to use for training autonomous cars to avoid accidents, because apparently they're particularly "confident" and "risky" compared with other Europeans.

The startup told the Evening Standard that "cars need to understand the full breadth of human behaviour before they’re ever going to be implemented into urban environments," and this includes how distracted they are and whether they're intending to walk into the road or not.

Alongside New York, Tokyo and Mumbai, footage of roads in London is being used to train the software to scan for humans and understand over 150 possible behaviours to predict what they're going to do.

The software will be able to tell whether a given pedestrian is paying attention or distracted – or staring at their phone – and what they're likely to do next, so the car can prepare to slow down or swerve out of the way if they're likely to collide.

Humanising Autonomy says of the scheme, "our pedestrian intent prediction platform makes autonomous vehicles safer and more efficient in urban environments. Using culture and context specific predictions, we capture the full complexity of human behaviour in urban environments around the world."

That culture-specific part is important, because people don't behave the same ways on roads around the world. An AI trained on driving in the UK would be absolutely baffled in India, and vice-versa.

Hopefully, this kind of endeavour will make autonomous vehicles safe for everyone – more so than human-driven cars, because goodness knows we don't have a clue what people are thinking sometimes when they dash out into the road.