The CEO and co-founder of a London-based AI firm says the tech must be focused on busy cities rather than "sunny California Highways".

Humanising Autonomy, which raised $5.3m in seed funding last June, has developed a way to predict pedestrian behaviour from inside a vehicle.

Maya Pindeus said the start-up is now focused on winning more contracts and licensing the technology.

Returning from CES, where the company demonstrated its technology on the streets of Las Vegas, Pindeus said the vision for the firm is to build a “global standard for human interaction with automated mobility”.

Without technology of this type, Pindeus predicts that autonomous vehicles will not be capable of navigating the tight streets of busy cities with safety and efficiency.

“If the automated revolution is to happen, which we believe it will, it has to function not just on a sunny California Highway but in real cities,” she told BusinessCloud.

“Our focus is what we believe to be the missing, crucial key in order to make automated vehicles at scale a reality.”

Founded in 2017, Pindeus co-founded the firm with Chief Product Officer Leslie Nooteboom and CTO Raunaq Bose.

It is already working with the likes of Transport for London, Arriva and Airbus to help better predict the actions of those potentially in the path of a vehicle driven by computer.

“Without being able to understand what pedestrians, cyclists, wheelchair users - the most vulnerable road users - in any city are doing, the technology remains unsafe, so there won’t be a large-scale adoption of automated vehicles in cities such as London and Tokyo,” she said.

Without it “the billions and billions that has been put into autonomous vehicle development won’t happen”.

The software is built to sit alongside existing navigation software in modern vehicles such as fully autonomous, semi-autonomous and fleet vehicles.

It doesn’t do the driving; instead, it plugs into any existing hardware and software to add a similar intuition to that of a human driver. 

Autonomous vehicles already have the ability to recognise when a pedestrian is in the road in order to avoid a collision, but Pindeus said Humanising Autonomy’s technology can predict what a pedestrian or cyclist will do next, which gives back crucial seconds to make alternate manoeuvres.

Screen

Humanising Autonomy's technology in action

“The vision for us is to build a global standard for human interaction with automated mobility,” said Pindeus.

“There needs to be an entity that defines what is safe and when it is safe enough to operate in pedestrianised environments.”

The company, which is based in the centre of London, is currently focused on use in densely populated cities.

The 25-strong team of behavioural psychologists and technologists use their local streets to test extensively in Soho and Covent Garden, as well as in cities such as Japan and the US. 

And while pedestrian safety is always the most pressing concern, she said that without the ability to predict pedestrians’ intent, traditional software remains overly cautious.

According to Pindeus, vehicles without this human-like intuition have to be very careful in busy environments “meaning it can’t really drive at normal speeds around people".

She adds: "Every time a car would see a person, it would have to slow down or stop completely. This leads to a lot of false positives.

“If a vehicle has to be so cautious around people that it actually doesn't even get to move properly, that means that there will likely not be any adoption of the technology.”

The firm’s AI-powered tech quantifies data points including a pedestrian’s intent to cross the road and if they are looking at their smartphone.

More than 150 of these factors are currently analysed, with more in development. These factors combine to give a car a ‘risk index’ of every pedestrian.

This ‘risk index’ also takes into account broader influences. A pedestrian or cyclist in one city may act differently at the roadside than those in another.

Pindeus explains that a “fusion between behavioural psychology and deep learning AI models” makes these differences possible to learn from and better prepare for.

“We quickly realised that cultural behaviours are different from location to location or environment to environment,” she said.

The technology is claimed to be capable of spotting dangers more than two seconds earlier than a human, and though the firm is keeping statistics close to its chest ahead of the publication of a whitepaper, Pindeus said the team counts every preventable accident avoided as a win.

“We don't provide the decision-making; that lies with whoever provides it, which is likely to be the manufacturer,” she explained.

“The liability and responsibility of the vehicle always lies with the OEM [the original equipment manufacturer]. We provide a likelihood of any event happening.

“Our approach is to say that this is the crucial technology that everybody manufacturer and every supplier requires, and there’s no way around it.”