How we’re helping our customers succeed with AI
21 November 2024
We recently announced major upgrades to Terrain Intelligence. But to benefit fully from groundbreaking deep tech, customers need to understand how to harness it – and this is where the unique composition of our team really shines.
“You can move mountains with the right talent.”
We recently announced that we have doubled inference speed and cut our model size by 75% in our 2024 Terrain Intelligence release, all while maintaining model prediction accuracy. How have we achieved this? The answer, according to our CTO Bill Lamey, can be summed up in two words: our people.
But for any start-up, building technology is just one part of the puzzle. Helping customers understand how to adapt to and adopt it is equally important and no less challenging.
“At Potential, we don’t just talk about how to build game-changing software,” confirms Bill. “We also talk about how we can help our customers understand how to plug this technology into their ecosystems as they move into this new era of software-defined vehicles.”
In the powersports and motorcycle sectors, many customers are in the early stages of offering onboard cameras and associated features. Automotive makers are often more familiar with cameras and compute capacity. What all these customers have in common, however, is that they are only just beginning to ask how AI and computer vision could benefit their products.
Terrain Intelligence 2024 model cuts inference speed by 55%, initialization time by 53%, and model size by 75% - and all without losing any accuracy
A first principles approach
“AI evolution is based on experimentation and iteration,” says Bill.
Much has been said about AI in the past few years, and it’s fair to say there’s been a lot of hype. In Bill’s experience, the possibilities that AI can open up are very real – and the pathway to developing and deploying AI can be reassuringly familiar.
Bill joined Potential after a 25-year career spent mapping ocean floors at Teledyne Caris and scaling IBM’s cyber security division. In 2016, Bill was leading an innovation cell at Teledyne Caris, the market-leader in remote sensing and measuring seabeds for safety-critical applications such as navigation charts. Under his direction, the team built and launched the first AI data-processing platform in the marine mapping industry.
Bill points out however that this step-change in that industry was less about the magic of the new and more about applying previous learnings in a systematic way.
“Advancements in computing power, network architecture and bandwidth in the early 2000s completely changed how we thought about solving computational problems ” says Bill. “There’s a parallel here. AI is also changing the way we think about the problems we want to solve and how to solve them. But the way we build teams to understand and develop AI technologies is similar in principle to what we were doing 25 years ago. Everything’s just moving faster!”
Bill Lamey, Potential CTO, thinks the pathway to developing and deploying AI can be reassuringly familiar
The challenges to AI adoption in automotive and powersports sectors
The major challenges to AI adoption we’re seeing with our customers can largely be broken down into the following areas:
1. Hardware. In powersports and motorcycles, some of our customers are working towards launching their first production vehicles with integrated onboard cameras. In automotive, customers are working towards true software-defined vehicles. Both present challenges to running pilot projects with Terrain Intelligence – how can we read from cameras if there aren’t any, and how can we present a right-sized footprint if future compute specifications haven’t yet been determined? We’ve been helping our customers unblock the path in different ways. For example, we have installed cameras and compute modules on test vehicles ourselves, taking care to use products that are likely to be used on production vehicles. In other cases, we’re advising on hardware specification to help customers make the right choices among a dizzying array of compute options.
2. Displays and controls. We’ve learned a lot on a project with one of the world’s leading motorcycle manufacturers. For example, how can we show what Terrain Intelligence is reading and calculating when the vehicle doesn’t have a large screen and no space to add one? Partly, this involves educating our customers on how to think about AI and its calculations, which often can’t be displayed on a screen. But in this case, we also designed a screen-less display module that we installed around the existing instrument binnacle, with a simple system to show riders data trigger points that can be read and understood even at high speed. In powersports, we built a display screen into the dash of our test side-by-side, complete with UX that clearly shows the rider which vehicle settings Terrain Intelligence is selecting. We’ve even fabricated 3D-printed dashboard surrounds and camera shrouds to keep test vehicles looking as close to OEM aesthetic as possible.
Part of a custom Potential install inside the cab of a side-by-side recreational off-road vehicle
3. End-to-end or deterministic. There is a present and ongoing debate in the automotive sector over the risks and opportunities offered by end-to-end AI (where the platform outputs decisions, such as changes to suspension, driveline, and drive mode settings) compared to deterministic AI (where the platform outputs information that can be used as sensor data to inform existing decision-making). This can be illustrated in the autonomous driving space, where Tesla and Wayve are following an end-to-end approach, while companies like Waymo and Cruise favor the deterministic approach. We have experience building technologies across both approaches, so we’ve been able to help our customers assess the merits of each.
4. Data. As Bill outlined in a recent article, data alone is not the answer to building reliable AI tools. No, the type of data, and verifying results, is what’s important. Many of our customers are only just beginning to explore how to capture, validate, store and utilise data. We’re already several years into this process. We’ve built many of our own automated tools based on our findings, and we conduct a regular test program, year-round, in many different vehicles (including motorcycles, side-by-sides, SUVs and pick-up trucks) and different environments, both off-road and on the road.
“Our job is to help our customers get to AI adoption faster than anyone else. ”
Why team culture and composition matter
Bill brought many vital lessons to Potential from his storied engineering career. Among them are these pearls of wisdom that form the foundation of our approach at Potential: “Listen to your engineers! Build quality early. Don’t wait to test. And blend highly technical and pragmatic approaches so that you can ‘fail fast’ – get to the answer quickly to determine if it’s right or wrong.”
At Potential’s HQ in Fredericton, New Brunswick, our AI experts sit right alongside engineers from mechanical, electronic, mechatronic and vehicle dynamics disciplines. Our product team, which feeds customer insights back into the technical team, is housed under the same roof. They all sit just a few feet from our fleet of test vehicles, which include motorcycles, side-by-sides, pick-up trucks and SUVs, as well as our own custom-built, quad-motor, all-electric test platform, P0.
We have our own off-road proving grounds at our HQ, access to hundreds of miles of trails, and we’ve already conducted testing programs in five countries and three continents, covering thousands of kilometers and many different off-road and on-road environments.
We also have team members stretched across Canada, the US and Europe, meaning we can reach our customers in person and better understand local environments. (Meet some of them here.)
This composition has proven attractive not just to our skilled engineers but also to industry luminaries like Scott Kunselman and Stacey Stewart, whom we’re proud to count among our strategic advisors.
“We’re building an honest, open culture with no wrong answers,” says Bill. “We test, we validate, we iterate, and we do it all again.”
Integrating AI tools such as Terrain Intelligence into vehicles to improve safety, performance and comfort is inevitable. We believe the brands that get there first will benefit from significant competitive advantage.
Bill concludes: “We see ourselves like rally co-drivers: our customers are in the driving seat, and our job is to help them get to AI adoption faster than anyone else. It’s an exciting place to be!”
Get in touch to talk about Terrain Intelligence, AI and the next frontiers in automotive and powersports technology. We’d love to hear from you.
