Humanising Autonomy uses behavioural psychology and computer algorithms to make cities safer for pedestrians and cyclists.

Using cameras and AI to protect vulnerable road users

Our Zenzic CAM Creator series continues with Raunaq Bose, co-founder of Humanising Autonomy.

Before establishing predictive artificial intelligence (AI) company Humanising Autonomy in 2017, Raunaq Bose studied mechanical engineering at Imperial College London and innovation design engineering at the Royal College of Art. Focusing on the safety of vulnerable road users, Humanising Autonomy aims to redefine how machines and people interact, making cities safer for pedestrians, cyclists and drivers alike.

RB: “Our model is a novel mix of behavioural psychology, deep learning and computer algorithms. We work with OEMs and Tier 1 suppliers on the cameras on vehicles, with the aftermarket on retrofitted dashcams, and also with infrastructure. Our software works on any camera system to look for interactions between vulnerable road users, vehicles and infrastructure in order to prevent accidents and near misses. While most AI companies use black box systems where you can’t understand why decisions are made, we set out to make our models more interpretable, ethically compliant and safety friendly.

“When it comes to questions like ‘Is this pedestrian going to cross the road?’, we look at body language and factors like how close they are to the edge of the pavement. We then put a percentage on the intention. Take distraction, for example, we cannot see it but we can infer it. Are they on the phone? Are they looking at the oncoming vehicle? Is their view blocked? These are all behaviours you can see and our algorithm identifies them and puts a numerical value on them. So we can say, for example, we’re 60% sure that this pedestrian is going to cross. This less binary approach is important in building trust – you don’t want lots of false positives, for the system to be pinging all the time.

“One of the main things we’re likely to see over the next decade is increased use of micromobility, such as cycling and e-scootering. At the same time you will see more communication between these different types of transportation, and also with vehicles and infrastructure. The whole point of ADAS is to augment the driver’s vision, to reduce blind spots and, if necessary, take control of the vehicle to avoid a shunt. Then there’s the EU agreement that by 2022 all buses and trucks must have safety features to detect and warn of vulnerable road users.

“We currently only look at what’s outside the vehicle, but with self-driving there will be monitoring of the cabin. In terms of privacy, we have a lot of documentation about our GNPR processes and how we safeguard our data. Importantly, we never identify people, for example, we never watch for a particular individual between camera streams. We look to the future with autonomous cars but for now we’re focused on what’s on the road today.”

For further info visit

Deadly driverless car crashes

Probably the highest profile fatal crash involving a driverless car occurred in Arizona in March 2018.

An Uber test car, in autonomous mode but with a safety driver, hit a 49-year-old homeless woman in the city of Tempe.

Elaine Herzberg was walking with a bicycle and not on a crossing. It was the first reported fatal crash in the US involving a self-driving vehicle and a pedestrian.

Fast forward nearly a year and the University of Michigan has unveiled a new project to predict pedestrian movements with greater accuracy.

“Prior work in this area has typically only looked at still images,” said Ram Vasudevan, assistant professor of mechanical engineering. “It wasn’t really concerned with how people move in three dimensions.”

By studying things like gait pace, foot placement and the symmetry of arms and legs, the team attempt to predict the future locations of one or several pedestrians up to 50 yards from the vehicle.

“If a pedestrian is playing with their phone, you know they’re distracted,” said Vasudevan. “Their pose and where they’re looking is telling you a lot about their level of attentiveness. It is also telling you a lot about what they’re capable of doing next.”

Previously, the most notorious driverless crash was also in the US, in 2016, when a Tesla Model S in autopilot mode smashed into a truck’s trailer, killing the car’s 40-year-old driver.

There have been numerous close shaves too.

Just last week in St. John’s, Canada, a driverless car reportedly set off at high-speed down a residential street, jumped a snow bank and slammed into a nearby garage.

Incredibly, no one was hurt. The Royal Newfoundland Constabulary (RNC) is investigating.

Cars of the past: who needs seatbelts?

In the week when 97-year-old Prince Philip did his best to put road safety back on the front pages – first by smashing his Land Rover into a blue Kia, then being spotted back behind the wheel but not wearing a seatbelt – the British Safety Council reminded us of the pioneering work of controversial founder, James Tye.

Tye (pictured) campaigned tirelessly for 25 years until wearing a seatbelt become a legal requirement in 1983, producing one of the first reports on the subject back in 1959. The Department for Transport estimates the humble harness now saves around 2,000 lives in the UK every year.

Matthew Holder, head of campaigns at the British Safety Council, said: “The times when critics of the seatbelt regulations accused the government of operating a nanny state and limiting their personal freedom and comfort are long gone.”

With multiple studies showing that 90% of accidents are caused by human error, how will we look back on reasons to fear driverless cars 50 years from now?