The future’s arrived

News stories are free to read. Click here for full access to all the features, articles and archive from only £8.99.

AI technology once seemed like the work of science fiction, but companies like Exeros Technologies are at the forefront of rolling the lifesaving systems out across the passenger transport industry. Peter Jackson learns more

If you hear the term AI – that’s artificial intelligence – what comes to mind? Perhaps its video games, predictive text or maybe even humanoid robots. Either way, it’s unlikely to be the number one bus. One company is setting out to change that perception: Exeros Technologies.

Are you enjoying this feature? Why not subscribe to continue reading?

Subscribe for just £10 a month with our annual print and digital offer, Or login if you are already a subscriber

By subscribing you will benefit from:

  • Operator & Supplier Profiles
  • Face-to-Face Interviews
  • Lastest News
  • Test Drives and Reviews
  • Legal Updates
  • Route Focus
  • Industry Insider Opinions
  • Passenger Perspective
  • Vehicle Launches
  • and much more!
[/wlm_nonmember][wlm_ismember] The firm has been working on AI systems of its own since 2014, as CTO Jay Biring told me: “We’ve been working on it for a while behind the scenes, but not really putting it to market. Obviously, there’s loads of work that goes into AI-based products, and there’s lots of learning required.”

That learning is machine learning – teaching the computer to recognise specific things and perform certain tasks in response. “Basically, you go out and place an algorithm – one that recognises driver fatigue for example – into a camera system,” explained Jay. “You then have to check and teach the machine, for example, what is yawning and what isn’t yawning, just somebody opening their mouth. Some of those things can be quite difficult. Imagine if you’re teaching a child what a yawn is – it’s kind of like that. The more you show the machine examples of what is and what isn’t a yawn, the more acute your algorithm becomes. There’s lots of retraining that’s needed before you can deploy an AI-based product, and that’s what we’ve been doing over the last two years. It takes time, but we’re there now.”

Exeros achieved a 100% accuracy read rate on cyclists when it tested the system on a London bus. EXEROS

Ahead of its time

I asked Jay how Exeros got its start in the coach and bus sector. “We first developed a camera that could go on the side of a bus, and it could accurately detect the shape of a cyclist in relation to the bus, calculating how far they were away from the bus,” he said. “At the time, the chipsets and sensor technology that was needed inside the camera was very expensive, and would have brought the cost of that camera up to about £1,200. What we learnt from that was really valuable. We had 100% accuracy read rate on cyclists when we tested the system out on a London bus – even at night and with poor visibility – and we realised that though it might not be a commercially available product at first because of the cost of the technology, that price would come down. Lo and behold, here we are today, some seven years later, and now that same camera – which we have commercially deployed – is around £100! It’s amazing how much the cost of silicone chips has diminished over time. It’s all about having an idea, and while technology might seem expensive now, it won’t be in five years’ time and it’ll be pretty commonplace. Things like reversing cameras on vehicles; I remember the first vehicles that came out with them on and thought how amazing they were. Now, pretty much every car has a reversing camera, at least as an optional extra.”

Since those early days, Jay and his team shifted their focus to other industries. That is until recently. “We’ve had such great successes in truck, so we looked at the bus market and asked operators ‘what are your challenges?’ They just wanted a single architecture in the vehicle that could control the CCTV, collision warnings and keep everything functioning as much as possible – and that’s what we went away and developed,” said Jay. “The camera I spoke about before is one module of that system, but the system itself is a scalable architecture for buses. This is something that hasn’t really been seen in the bus industry – it’s always been a separate supplier for camera, a separate supplier for your advertising platform… that makes the transition to new technology very difficult for operators. If they had something that was scalable and modular, we’d start to see rapid improvement in technology deployment – this is what the industry is in dire need of.”

AI-based technologies

Speaking about the rise in popularity of AI tech, Jay continued: “These kinds of technologies will become commonplace – they’re already in our cars, watching the road ahead. If the front-mounted camera on a car spots an obstacle in the road and the driver isn’t paying attention, the car will apply the brakes. It saves lives, and this is the same technology we’re applying here. The technology is there now, so we should leverage it as an industry and embrace it – it will make sure a bus driver and their passengers can go home to their families each day. It will make sure everyone on the bus is protected by an additional layer, rather than relying solely on the driver. There’s a lot of onus placed on the driver currently to keep passengers safe, so if we can assist them in their job using technology, why wouldn’t we? 94% of accidents are caused by human error – if we give drivers a two-second early warning of an impending collision, we can reduce that down to 1%. We’re trying to design our systems to pre-empt things as much as we can, the goal being to warn drivers two seconds in advance of an accident. We’re probably at about 1.5-1.8 at the moment, it depends on the event.”

The Driver State Monitor uses a small dash mounted camera. EXEROS

So, what systems does Exeros offer and how do they work?

Cyclist Detection System: This warns the driver if a cyclist is too close to the vehicle and triggers an audio alarm outside the vehicle to warn the cyclist that they’re in danger. “The communication between the vehicle and the cyclist is imperative when it comes to saving lives,” said Jay.

Driver State Monitor: A small dash-mounted camera that will alert the driver if they’re falling asleep or distracted by using a mobile phone, for example. It will also alert them if they’re not wearing their seatbelt or smoking behind the wheel. ”It’s not really about sending images off to a fleet manager,” said Jay, “it’s all about having a second set of eyes to keep the driver and the passengers safe.”

Forward-facing camera: This detects pedestrians about to step out in front of vehicle, whether the vehicle is tailgating or veering out of lane unexpectedly and detects road signs. If the driver is looking at traffic lights and creeping forward, or looking right at a roundabout and a pedestrian steps out in front of the vehicle, the system will alert the driver.

Right now, the systems don’t apply the brakes, they just warn the driver. “These drivers are already highly trained – they’re some of the best drivers in the country, hands down,” said Jay. “So we’re giving them assistance first, but we are talking to a couple of bus manufacturers now about using the sensor technology for automatic brake application and assisted steering as well. The challenge there is injecting CAN code into the vehicle and looking at control mechanisms – sometimes the vehicles don’t have those interfaces – the manufacturers are working hard to bring that up to speed.”

Somerset Passenger Solution is utilising Exeros AI passenger counting systems on its fleet. EXEROS

Passenger counting system

Separate to those safety systems, Exeros has also built an AI passenger counting platform for the Hinkley Point Power Station transport provider, Somerset Passenger Solutions – an operator which transports around 6,000 passengers daily to the site aboard 150 vehicles. Because of Covid-19, Hinkley had to operate its buses at 50% capacity, which is gradually increasing now restrictions are being eased, so the operator wanted to be able to monitor precisely the number of people on the bus as it doesn’t have a ticketing system.

“Their technology challenge was that they don’t have a set route structure; vehicles do different routes each day and new stops can be introduced at relatively short notice,” said Jay. “We used a time-of-flight camera, which is very accurate at detecting the shapes of people as they get on and off the bus, and we’re getting a 98-100% read rate in terms of counting people coming on and off the bus. Generally, it’s much closer to 100% than 98%. As with anything machine-learning based, we’re still teaching it, but we’ll get it there very quickly.

“We had to build the AI platform so that it could automatically ascertain which route the bus had taken and how many people had got on and off at each stop. The AI system determines what route the vehicle took without any input from a scheduling front-end. This means they don’t have to integrate their scheduling system into our platform – it works independently and gives them everything they need. There are a lot of use cases for this system and it can be deployed to any bus operator in the UK. And because the AI is already there, they don’t have to do any front-end coding on it. All they have to do is mark the stops of each route on a map, and the system will learn which route the vehicle is on by itself within a couple of days.”

The system can link in to an operator’s app and tell passengers how many seats are available on each bus.

Eliminating bridge strikes

Exeros is currently working on a low bridge detection system, utilising the same front camera as used on its CCTV system – meaning no new equipment will have to be fitted to the bus with this tech already in place. The AI is being trained currently to visually recognise and measure low bridges. So far, Exeros has managed to get the system down to a sub-10cm accuracy level on early trials.

“Bridge strikes can cost a company £1-1.5m just in the damage to the bridge,” said Jay. “It affects the rail system, services like gas and water and there’s a time cost as well to fix it – all of these costs rack up very quickly. And that’s if there isn’t any loss of life. We’re working with a new algorithm which is designed to read the road signs pointing out the size of the bridge on the approach, and cross check that against the measurement of the bridge once the camera can spot it. The system will warn the driver though before the vehicle even gets to the bridge, as soon as it reads the signs. Before that even, we’re going to set up geozones, so we’ll let the driver know they’re approaching a bridge before they see the signs. So it’s a four-stage safety system, and we’re very close to completing it now.”

Looking ahead

In terms of the future integration of AI into the coach and bus sector, I asked Jay whether he thought fully autonomous buses would become widespread. “The bus market hasn’t yet been exposed to what can be done with AI,” he said. “Sometimes, when we showcase our AI technology on the bus and some of the smart cameras, bus operators tell us they didn’t think that kind of tech would hit the market for another 10-15 years. That has traditionally been the case; a BMW 7-Series for example will have the latest tech on it and do all of these wonderful things, but it’s not carrying 50 people so its risk profile is lower. Why aren’t we putting this technology on buses and HGVs driving around city centres? Because actually, they need it more – if they do have an accident, it’s more severe. I think there will be a turning point, and we’re at the very beginning of that transition. Let’s say, for example, that in the industry today we’re relying 80% on humans and 20% on computer assistance. I think there will be a slow transition over the years, dropping to 60-40 and then 50-50, before switching to 40-60 in favour of technology. But I think it will end at 20-80. Humans will still be involved in the movement of the vehicle, but the system will take on 80% of the processing workload. I think, especially if you look at private vehicles, there’s a lot of ‘wasted concentration.’ We sit in a car, holding a steering wheel, marginally adjusting it on the motorway and gently feathering a pedal, just being perceptive. But we’re occupying so much of our time in that scenario – could we not be doing something else with that time? I think the same will apply to the PCV market. The challenge for the industry today to get there is that we need a lot more training data. We haven’t deployed enough early technology yet to harvest that data and feed back for future improvement. We’ve got an answer for this in the pipeline, which is coming out pretty soon. It’s all about sharing data with other companies developing systems, creating an open platform for collaboration so that we can save more lives.”

Tech companies are also rapidly developing ways for AI-equipped vehicles to talk to each other, as well as things like intelligent street lamps to spot obstacles and communicate the information to passing vehicles. “5G is one of the big requirements for this to happen – we need faster and lower bandwidth data communications,” said Jay. “We’re not using megabytes worth of data to send information from a lamppost to a vehicle, we’re using a few kilobytes. This is where the AI tech companies are focusing – how do we slim down the amount of data we use? We have to make the data so accurate that it can be slimmed down as much as possible while still being 100% reliable.”