The differential of the vehicle allows left and right wheels to turn at different speeds.
This is just one of the first and most basic examples of the challenges the TomTom engineers faced. Besides wrestling with code iterations and troubleshooting edge cases, the engineers had to develop an understanding of how the driving dynamics affect how a vehicle changes direction in the most detailed way.
How fast a vehicle is traveling, the type of turn it’s making, the scenario in which it’s driving and the speed it steers all affect the geometry of the turn it will make. The issue is that all these different turns are represented by different shapes and geometries. What’s more, vehicles tend to have more than one axle, which also affects how it changes direction. If the model the TomTom engineers were developing only accounted for one axle, mapped turns wouldn’t always be accurately reflected.
Think about when you change lanes on a highway or take the exit ramp. These types of maneuvers are quite different from turning a corner in a city. Generally speaking, in city driving, when making turns around corners, the turns can be mapped by parts of a circle. However, on multi-lane highways and slip roads, turns display a different geometry.
“This effect is more pronounced in some situations”, Dr Bednarz adds. “On multi lane roads, slip roads and highway entries and exits, roads designed to maintain a high flow of vehicles, the turns vehicles tend to make are represented by parts of clothoids [complex curve shapes] rather than parts of circles.”
If these aren’t accounted for, the system would incorrectly adjust the trajectory of the vehicle and its calculated position could be incorrect.
“If you drive and steer in a two-axle vehicle, but the turn being mapped is based on the geometric model for a single-axle system, the mapped turn will be greater than the actual turn made, unless the turn represents part of a circle.”
If you were wondering what a clothoid, also known as a Euler spiral, is, it’s a type of curve, the curvature of which changes along its length. They’re used a lot in railway design to transition between two parallel linear tracks. You’ll almost certainly have seen them before, even if you didn’t know the name.
Uncovering these dynamics can only happen during testing. Rather than having to go out and drive in the real world to test the effectiveness of the model, the team can simulate an entire day’s worth of driving within a few seconds. They can then compare the model’s performance against a known truth by overlaying the simulated vehicle’s trajectory on a real map. This allows them to spot any errors where the track doesn’t follow the road. After this, the process of iterative improvement can begin again.
With every code revision, the model’s accuracy has improved, but it’s still in constant development. The team has thousands of kilometers worth of historical drive data which it can test new iterations of the model against.
Dr Bednarz believes the model has potential for further development and will be on par with the performance of the gyroscope on a chip. Speaking with him, I get the sense he’s being humble and that developing this solution is a pursuit of perfection, and for Bednarz and his colleagues, the work will never be completely finished. They will always be looking for ways to improve their model.
With good reason, too, because it could have a dramatic impact on the automotive industry.
Cost savings and cheaper navigation
While the chip shortage gave the motivation necessary for Bednarz, Leclare et al. to develop their system and help carmakers navigate their way around supply chain issues, it will also open new opportunities for carmakers to forgo the use of the gyroscope chip in markets where it’s not really needed.
The algorithm-based model isn’t just a band-aid solution, it could very easily replace gyroscope chips entirely.
In some countries, particularly in South America, there isn’t as much need for gyroscope chips as there is in places like Europe and the US. There aren’t many road tunnels or highly built-up areas which can negatively affect GNSS signals.