Tech News

Wednesday, 2 November 2016 | MYT 8:30 AM

Swift Navigation aims for pinpoint GPS to guide autonomous cars

Pinpoint accuracy: A filepic of a Google car with the video captured by the vehicle (inset) coupled with the same street scene as the data is visualised by the car. Swift said its receivers would benefit not just fully autonomous cars, called Level 5, but also vehicles at Level 4 (where the vehicle is autonomous except in difficult situations such as severe weather) and Level 3 (vehicles that are autonomous in limited, known environments). — Reuters

Pinpoint accuracy: A filepic of a Google car with the video captured by the vehicle (inset) coupled with the same street scene as the data is visualised by the car. Swift said its receivers would benefit not just fully autonomous cars, called Level 5, but also vehicles at Level 4 (where the vehicle is autonomous except in difficult situations such as severe weather) and Level 3 (vehicles that are autonomous in limited, known environments). — Reuters

Most GPS receivers are accurate within a 5- to 15-foot radius – the size of an above-ground swimming pool. That's enough to find the nearest Safeway or Starbucks, but would be a disaster for self-driving cars or drones. 

To address that market, San Francisco's Swift Navigation is developing GPS receivers that are accurate within less than an inch – the diameter of a quarter. 

"Our purpose is to enable a future autonomous vehicle to understand and navigate the world," said Swift CEO Tim Harris, 30, an attorney, who co-founded the company with engineers Fergus Noble, 28, and Colin Beighley, 27. 

"GPS fused with inertial measurement units (which detect direction and acceleration) provides the only absolute position, time and velocity" information. 

Swift's view is that self-driving cars need five complementary sensors to perceive the world: GPS, cameras, radar, laser devices called lidar, and inertial measurement units. 

"Everything else helps the car focus on 'seeing' the things around it, but there is very little understanding of where the car is, other than GPS," said Rob Hayes, a partner at First Round Capital, which invested in Swift – and was among the earliest investors in Uber. 

Swift, which crowdfunded its first receiver on Kickstarter in 2013, now has US$14mil (RM58.55mil) in VC backing. Swift's precision GPS provides "deep, deep technology in very uncertain environments," Hayes said. 

The scores of carmakers and tech companies now racing to develop autonomous vehicles are secretive about their technology. None has discussed using precision GPS, although most mention cameras, radar and lidar. 

"There are two schools of thought" among developers, said Raj Rajkumar, a Carnegie Mellon engineering professor and expert on autonomous cars. "One, exemplified by Google, is that you don't need GPS; instead it does intense, fine-grained data collection about the roads and matches data from its onboard sensors with that data. The other school, exemplified by many carmakers, is that GPS is useful. We don't know which will win the day." 

Existing high-precision GPS receivers cost upward of US$10,000 (RM41,828), making them impractical for mass use. Swift says its systems, now US$600 (RM2,509), will get even cheaper over time and as more are manufactured. 

Here's how Swift's Piksi Multi, due early next year, works: Like all GPS receivers, it acquires signals from up to 12 of the satellites orbiting the Earth at an altitude of 12,550 miles. But the ionosphere can distort those signals. So Piksi uses either a cellular or radio connection to communicate with a network of ground-based stations for extra calibration. (For limited-range use, such as a self-driving tractor, Swift provides a base station that the user installs.) 

It's also a multiconstellation receiver, so it can acquire signals not just from the 31 US satellites, but from the satellite constellations owned by Russia, Europe and China. And it's multiband, meaning that it picks up signals on two different frequencies from each satellite. Finally, it can measure the actual wavelength of the radio waves emitted by the GPS satellites for extra fine-tuning. 

And it can do its calculations in 3 seconds, fast enough to help cars stay in their lanes, said Noble, Swift's chief technology officer. 

Patrick Connolly, principal analyst at ABI Research and an expert on location technology, views Swift as an exciting startup with the potential to change the industry. "We expect precision GPS to be a common component of any fully autonomous vehicles," he said. 

But there's an impediment: The need for base stations. "Getting that network is a big barrier," he said. "It could be tens or hundreds of millions of dollars" for worldwide coverage. 

Such networks do exist already. The most-comprehensive, called CORS, the Continuously Operating Reference Stations network, has good coverage in urban areas, such as the East and West coasts of the US, and scantier coverage in rural areas, such as the Midwest. Other players with much deeper pockets than Swift are looking to enhance the ground networks, Connolly said. 

All the sensors for driverless cars have some shortcomings. Cameras, radar and lidar may falter in inclement weather. GPS doesn't work in the concrete canyons of dense cities, or in tunnels or parking garages. 

"If it's snowing and there are no visible lane markers, it's very hard for an autonomous car to know where to be on the road unless it has a GPS system with hyper-accuracy," said Karl Brauer, executive publisher of Kelley Blue Book. 

Conversely, other sensors might have to step in when a car lacks a clear line of sight to the sky, disabling its GPS. 

Swift said it had almost 20 customers in the car universe, all of which required agreements that it not name them. Many of those might just be trying out Swift's technology, rather than fully committing to it. Of the startup's 2,000 customers, half are hobbyists and half are original equipment manufacturers, Harris said. 

Swift said its receivers would benefit not just fully autonomous cars, called Level 5, but also vehicles at Level 4 (where the vehicle is autonomous except in difficult situations such as severe weather) and Level 3 (vehicles that are autonomous in limited, known environments). 

With some 220 million connected cars predicted to be on the roads by 2020, that's a gargantuan market. 

"We'll be ready to provide hundreds of thousands, or millions of units" when the industry is ready, Harris said. — San Francisco Chronicle/Tribune News Service

advertisement

Powered by

advertisement

advertisement