An artificial intelligence camera system that can give bus drivers early warning of potential dangers is to be tested on public routes with a Transport for London-approved operator. Camden-based Humanising Autonomy say their ‘deep learning’ software can predict manoeuvres from vulnerable road users and identify hazards two seconds quicker than a human.
Speaking to the London Evening Standard, Raunaq Bose, the start-up’s co-founder, said: “Intent prediction shows not only where the cyclist is but if they are likely to move in front of your path. It helps with blind spots and people coming out into the road at random times.
“We’re able to tell if someone is aware of the vehicle or not, if they’re distracted, whereas current systems can just look at their position in the road. It pays attention to the subtle, nuanced motion and body language of pedestrians and cyclists.”
Every person picked up by the vehicle’s cameras is assigned a ‘risk index’ with pedestrians and cyclists classified as ‘vulnerable road users’ in need of protection.
Certain movement categories, such as ‘intention to cross’, will be flagged to the driver, triggering an alert.
Humanising Autonomy technology from Humanising Autonomy on Vimeo.
After two years studying tens of thousands of hours of human reactions from dashcam videos and test vehicle footage, the software also distinguishes between how people are likely to react in different areas.
“Tests found that in central London, where there are many more pedestrians and cyclists on the street, they are less likely to listen to the rules,” said Bose.
Earlier this week, a white paper produced by road safety charity IAM RoadSmart on driver education warned that businesses shouldn’t rely on technology alone to make their staff drive better.
“Technology is often relied upon to provide a solution to poor driving. In itself however, it rarely influences driver behaviour or attitudes,” said business development director, Tony Greenidge.
Stressing the importance of improving the skills and behaviour of drivers, he added: “There is still nothing to replace the direct educational feedback delivered by a professionally qualified trainer who is actually experiencing what is going on around them. They help change the way you think by linking their advice to a real and live example.”
Add new comment
12 comments
"Sorry Your Honour, I ran down the group of cyclists becasue my AI didn't recognise them and warn me"
No problems Bus Driver, you're found not guilty.
Reduce the demands on the drivers to keep to schedule and a lot of the stress and therefore urgency would be removed. I'm not saying that that is the only reason why some bus drivers drive so badly around cyclists but it would remove some of their beliefs that some of them have.
So AI can read our minds / see the future? Scary...
OK, let's give the bus driver yet more stimuli to process when the actual problem is they're up against it trying to keep to timetables and don't give a toss about cyclists getting in the way?
It's a great 'test problem' for AI I suppose and I'd assume it will collect data that will ultimately help improve safety of driverless cars so that's a big benefit for everyone, but in the short term I can't see this reducing accidents at all.
You might be right. But how do we find out? Do we just go with your gut instinct or do we set up an experiment to test the proposition?
Hmmm... what would a scientist do?
You'd hope that sticking to timetables, which isn't reasonably within the driver's control, would be the last of their priorities. Quite simply, it takes as long to get there as it takes to get there.
I think that the bus driver usually gets in trouble if they run too far behind the posted schedule...
You'd think, yes. But I'm afraid I have bad news. They're really pressured to 'maintain headway'.
“Tests found that in central London, where there are many more pedestrians and cyclists on the street, they are less likely to listen to the rules,” said Bose.
Where's the evidence then you snake-oil salesman?
TBH, it's the majority of road users in London, traffic would flow a lot better if everyone followed the rules... and as a side benefit, you'd massively reduce the KSI figures.
You'd have significantly fewer KSIs if more drivers of motorised vehicles observed the rules, but that's not what she said.
Hard to unpack what that means.
Is 'traffic flow' the most important consideration? And what constitutes 'traffic'? You'd massively reduce the KSI figures if you got rid of most of the motorised vehicles, of course, and almost certainly hugely improve traffic flow, if by that you mean the movement of people rather than vehicles.
And the quoted statement is pretty confusing as well. Maybe they 'listen' to the rules, but find them far too much orientated to the needs of the motorised and hard to follow for everyone else? Maybe they just don't agree with the rules? Perhaps it's the rules that are wrong, not the people?