AI and the Future of Cycling
Mike McGuireWhat the booking technology means to various parts of our sport
By Jake Williams @ Canadian Cycling Magazine
Booting up your training software and shopping for an e-bike have something in common these days. As consumers and cyclists in Canada, we’re being pitched AI, or artificial intelligence, and everything with a microchip now seems to boast that it’s the smartest version of itself with cutting-edge innovation. In reality, what AI means is many different things. Think of AI as an umbrella term encompassing areas of study like robotics and machine learning, programs that analyze massive amounts of data with limited human intervention. Netflix’s ability to recommend your next binge is as much AI as ChatGPT, the large language model chatbot capable of recounting most of the internet.
AI’s mainstream presence in Canada has increased in recent years, and the industry is projected to grow from $6.5 billion today to $28.2 billion by 2028, according to Innovation, Science and Economic Development Canada. The projected growth rate of almost 34 per cent is helped by the $2 billion investment set out in the federal government’s 2024 budget. With that in mind, the use and adoption of AI is still new for businesses in Canada. Statistics Canada reports only six per cent of businesses made use of AI in producing goods and delivering services in 2023.
Narrowing in on the cycling industry, AI is beginning to influence the way we train as athletes, our safety and efficiency on a bicycle, and even the way bicycles are designed. Of course, some companies are using the buzz of certain AI tools to capitalize on sales. Urtopia, a U.S. e-bike startup, has claimed to have produced the world’s first smart e-bike. Named the GPTS, the e-bike launched earlier this year and features all of the modern e-bike trappings: a Bluetooth speaker, app, and GPS integration. What’s new is that Urtopia has plopped ChatGPT into the bike’s head unit. “This bike can introduce itself,” reads the company’s website, featuring a video of a rider having a clunky conversation while on a bike path. The cyclist tells his bike he’s getting thirsty. In response, the chatbot recommends adding a coffee stop to the route. After the rider says he wants something more exciting, the monotone voice recommends a whiskey bar. The rider starts to turn around and shouts, “You get me, bro!”
While AI is a lot more than your virtual drinking buddy, many Canadians might not know any better. Only 16 per cent of Canadians have ever used an AI tool like a chatbot, according to the annual survey of the Canadian Internet Registration Authority (CIRA).
“A lot of people think AI is ChatGPT,” says Armando Mastracci, a graduate of engineering science at the University of Toronto and founder of Xert, an online training platform for cyclists. “In ChatGPT, you can ask for a training plan, and it will give you a bunch of workouts. The way it finds those workouts and how it does that, however, is very different from creating a neural network on a machine learning platform,” Mastracci says. He launched Xert in 2015 after pivoting away from working on an automatic shifting platform. His fascination with analyzing cyclists’ data came from working with his own—recorded from a recumbent he used for training during the winter months in Toronto. After creating mathematical models to assess his data, he saw patterns connected with athletic improvement. The next step was the software that would analyze riders’ historical data without human help, eventually morphing into Xert.
Xert’s latest tool, Forecast AI, goes a step beyond recommending a rider’s next workout based on their historical data. The program, which is still in beta, allows you to set an end goal of improvement, whether that’s a four-minute maximum power or a higher functional threshold power, that’s up to you. While scraping your historical data, the program will give you a complete work-back plan to achieve your goal. Mastracci insists he was finally ready to use the term AI based on the sheer amount of computational work done in the back end, data crunching that just wasn’t possible back when he was still riding his recumbent. As training plans like Xert increase their ability to hoover and analyze data, Mastracci sees a future beyond better FTP numbers. “When designing a race bike, you can put it in the wind tunnel and optimize its aerodynamic position and yaw angles, but racing isn’t just yaw angles,” he says. With the growth of glucose monitors and even better heart-rate data emerging, he hints at a more holistic training platform in the future.
As many Xert users were stuck on their indoor trainers through the early days of the COVID-19 pandemic, Clark Haynes was in early development of his so-called “pandemic passion project.” Haynes has spent the past 20 years working on advanced robotics and autonomous vehicles, while also being a bike commuter in Pittsburgh. In 2022, he founded VeloAI, a technology company specializing in safer streets. Its first product, the Copilot, is similar to a rear radar, but it also works in conjunction with an AI-powered camera. The unit is about the same size as a Garmin Varia radar, and allows you to track exactly how close and how aggressively an oncoming car will pass you. The unit connects to an app to give you a detailed visual representation of what’s behind. Up until VeloAI became a full-time job for Haynes, the robotics expert worked at Uber and then Aurora, an autonomous trucking company. He was in charge of what is referred to as prediction.
“We had to understand all of the crazy stuff that the other drivers, pedestrians or cyclists in the world were doing, and kind of relate that back to the autonomous vehicle. It just kind of clicked like, oh, what I really want is that prediction system on my bicycle, so that I can know what’s going on around me without having to constantly be watching myself,” Haynes says.
He explains that radar is good at telling you one or two things, but not everything. “If a car is approaching you, the radar can’t tell if it’s going to hit you or not,” he says. The Copilot’s camera, however, can track movement more precisely. Crucially, it aims to predict when a driver is operating erratically or driving too closely, a system that took a lot of data to be able to work. Haynes echoes Mastracci. “The AI is just a bunch of algorithms that are really, really well purposed to copying their data and transcribing their data, and interpolating from the data, but it all comes back to the quality of the data,” Haynes says. “If you don’t start with a good data set, you won’t have good AI.”
Haynes and his team spent numerous hours recording as they biked the streets of Pittsburgh, collecting the building blocks that help the Copilot’s AI prediction. Haynes explains that while camera data from a car would be useful, the nature of how a bicycle travels was integral to the success of the algorithms. A car travels smoothly on four wheels and suspension. A bicycle has more freedom of movement and is typically a rougher ride, so tailoring the Copilot’s technology to a rougher environment was critical to its accuracy, says Haynes.
In addition to Copilot’s benefits to bike riders, Haynes has started working with municipalities on planning and infrastructure using his invention. By capturing a high volume of riders on different routes, Haynes and his team can analyze the data and make recommendations to help identify particularly dangerous areas. “As we know, there isn’t a limitless budget to plop down bike lanes everywhere,” says Haynes, who is focusing on trying to maximize the impact of infrastructure. “The reality is that bike lights can only go so far when it comes to safe streets.”
While AI-powered cameras look to improve riders’ safety, another group from Ottawa is focused on riders’ comfort and efficiency. Jesse Jarjour launched MyVeloFit in 2020. The system lets you perform your own bike fit at home with only a smartphone and a bike, of course. Jarjour was a full-time fitter up until he launched the online fit tool. The idea grew from clients wanting more frequent bike fits without the extra costs and time. That got him thinking. With his background in economics and data analysis, he coded the first version of MyVeloFit, which continues to grow. “We’re up to about 1,200 fits a week,” he says. It’s a figure that is well above what he used to do in a single year. Instead of visiting your local shop or studio, you upload a side profile video of yourself riding on a stationary trainer. Within minutes, the program analyzes and recommends changes to your position. This process sets its users back about $50, which is roughly one quarter of what you’ll spend on a basic in-person bike fit.
Technically speaking, MyVeloFit uses a type of AI called computer vision. Developed in large part for autonomous vehicles, it allows Jarjour’s machine learning algorithms to recognize the difference between an ankle and an elbow. To assess a rider’s position, the software uses pose estimation, a task that relies on machine learning to identify and track each part of the body.
With more than 200,000 videos, Jarjour thinks his company might just have the biggest database of people riding bikes out there, which has value in and of itself. “So we’re kind of feeding that back into the system to start to get better ideas of how people move on bikes, and then start to look at new technologies based on that,” Jarjour says.
Of course, MyVeloFit raises similar questions as Xert. If we’ve limited the need for human intervention, do we need the human at all? In Jarfour’s case, he’s surprised. He expected more negative feedback from fellow bike fitters. In fact, the opposite Is true. Bike fitters around the world are using MyVeloFlt as a tool as It’s more accesslble and less expensive than large-scale motion capture systems, says Jarjour. Of course, the system Isn’t perfect, Its creator admits. Jarjour spends a fair amount of time chatting with clients about their position. “But in the future, that’s maybe something a chatbot can handle,” he says.
Mastracci doesn’t foresee a chatbot replacing a coach in cycling. “People certainly want questions answered, but I don’t see a bot being able to provide the insight and guidance in the personal context of an individual athlete,” Mastracci says. He sees Xert as a coach’s tool, removing the time-consuming computational aspect and giving more time to focus on the personal relationship.
Streamlining these time-consuming processes is often at the center of the decision to use AI. Roughly five years ago, component manufacturer SRAM partnered with software company Autodesk to create a prototype crankarm using a method called generative design. Instead of beginning with the drawings of an engineer or designer, the user sets the parameters of the end result. Then, the computer goes to work, pulling from massive datasets of existing and learned designs, creating new patterns. In a matter of hours, the software can produce thousands of potential options.
Following the crankarm project, SRAM has shifted to evaluating where to apply AI and machine learning. The biggest challenge for the company is deciding where to prioritize AI implementation in their products. SRAM is one of the larger cycling brands to go public with its work with AI. Shimano, for example, filed a patent this past January for a control device for a telescopic mechanism that receives output from “a learning model.” The patent includes the method for creating the learning model, the learning model itself, and a computer-readable storage medium. It all sounds AI-like. Shimano, however, doesn’t provide comments on “products whether they are in development or not,” including this patent. Currently, Trek is not open to discussing any AI product plans. Giant said it didn’t “have any immediate plans to incorporate AI into product development, design, or marketing.” If any of these companies are in fact working with AI in some form, it seems like they’re treating that technology like any bike or component: they’ll announce it when it’s ready to launch.
From smarter training tools to safety devices that predict danger, as well as the design of components, AI is reshaping the way we ride. As with any emerging technology, its thoughtful application is just as important as its data and capabilities. While AI can streamline processes and offer new insights, the human element—whether it’s a coach, a bike fitter, or a designer—remains crucial. AI may set the pace, but it’s the cyclists who will determine the direction.
