I met Tim Harris
, CEO of Swift Navigation
, several years ago just as I was beginning in startups. His leadership style is no small part of why Swift has earned a reputation as a fantastic place to work.
But it’s also a company overcoming big challenges in navigation — getting super-high precision out of GPS. Their success at generating near-centimeter accurate position data has major implications for autonomous systems where just knowing exactly where you are is a big deal.
We caught up with Tim to see what Swift has been up to.
How did you get started working on precision GPS?
My co-founders were working on airborne wind turbines, which were large rigid kites that would fly up to a kilometer above the Earth generating power. They were looking to use precision GPS to control these things. What we realized was that the market needed precise, absolute localization for autonomous systems.
Why do autonomous vehicles need such precise localization?
Autonomous vehicles need to do two tasks. First, obstacle detection, which amounts to “don’t hit anything. Second, precise and robust localization, which consists of relative localization using LIDAR or cameras along with computer vision and absolute positioning on a high definition map. Relative positioning requires extensive learning of an environment, so if you extend operations to a wider area, you need to rely heavily on absolute localization, and that’s the layer we provide.
So how does SWIFT Navigation address this?
Swift Nav removes the errors inherent in GPS by mapping sources of environmental error – the largest of which is the atmosphere itself. We monitor and model the impact of the ionosphere and troposphere in real time. As the ionosphere gets hit by different sources of space radiation, it gets electrically charged, which creates a delay in the signals passing through it. That’s what we try to correct for.
With all of those errors, you’re looking at 3 to 5 meters of ground accuracy. It can get a lot worse if you suffer multi-path signal reflection off of objects in cities, such as buildings. After correcting the errors, we get down to 2 centimeters of precision.
Early on, there were some technological developments that let us build our system: low cost, high power FPGAs and the availability of certain RF chip sets, for instance. However, since we’re taking a large chunk out of the price to make the technology widely available, we’re betting on there being enough larger applications to support this. It’s clear that automotive is moving toward full autonomy, but the numbers are low.
Automotive is moving towards full autonomy, but the numbers are still small. Our current focus is on semi-autonomous systems. ‘Fly or drive by wire’ has been around for years. That’s the market we’re addressing first. We’re already serving thousands of customers with applications that are coming out in volume over the next few years. We’re not talking about cars without steering wheels; we’re talking about cars with steering wheels that help drivers be a little better.
Do you see this level of precision GPS becoming more common in consumer applications?
It’s definitely possible, but there are systems integrations trade-offs. For instance, if you want to get this technology into a cell phone, you have to solve an antenna problem. If you’re trying to listen to a wide range of RF on a small element, you can either add more antenna, or make more of that antenna available for GPS. It’s a sandbox, and if you’re going to give us more space to build a castle, you have to add antenna cost or take that away from cellular, Bluetooth, and WiFi. It’s just a matter of when the value is high enough that you’re willing to make those trade-offs.