Innovation

|

07 Dec 2017

Boosting GNSS with sensor fusion

Boosting GNSS with sensor fusion

Two cars driving into a garage

Has your car’s GPS ever led you astray? After following it blindly for an hour, it suddenly dawns on you that you’re heading in the wrong direction. Your immediate instinct is to blame the device, even though it’s much more likely that you misinterpreted its instructions or mistyped your destination. Either way, you’re fuming behind the wheel and not an inch closer to where you want to go.

That said, your GPS isn’t infallible. Almost invariably, situations in which the GPS is the culprit trace back to something obstructing the weak signals emitted from orbiting navigation satellites. Urban canyons can deflect the signals, leading to significant errors in your position reading. Tunnels cut them off entirely.

These errors may seem trivial today – a hassle at worst. But as our cars get smarter and ever more autonomous, they could become consequential. You might be overcharged by pay-as-you-go urban road tolling systems that reach their limits in dense urban environments. Even worse, your autonomous car could misinterpret its location, apply the wrong speed limit, and take off, endangering you, your fellow passengers, and those around you.

Staying on the map with automotive dead reckoning

To solve these problems, u-blox developed its automotive dead reckoning solution, which it first integrated into some of its Global Navigation Satellite System (GNSS) receiver modules in 2003. Automotive dead reckoning uses data generated by sensors on the vehicle to improve GNSS accuracy in otherwise difficult environments, extending its reach deep into tunnels, multi-level car parks, and narrow urban canyons.

In the very best case, vehicles are equipped with the full suite of sensors required for high-performance and easy-to-use 3D dead reckoning. One such setup involves a three-axis gyroscope to detect changes in the vehicle’s angular orientation, a three-axis accelerometer to determine changes in speed, and a wheel tick sensor to measure the distance covered by the vehicle on the ground. Often, however, manufacturers opt to reduce costs by outfitting their vehicles with a smaller sensor set.

Fortunately, this doesn’t make dead reckoning a dead end. Sure, less sensors typically mean less accuracy. But with a bit of ingenuity, combinations of sensors can be used to recover some of the missing information.

 As a simple example, imagine that your car, equipped with a wheel tick sensor, enters a straight, flat tunnel. As long as the sensor is properly calibrated, it will continue to provide the vehicle’s precise position while GNSS reception is interrupted. If, instead, the tunnel goes downhill, the reported position will be off because the distance on the ground will exceed its horizontal projection. Others have solved this problem by determining the slope of the road using a single accelerometer and a bit of mathematical juggling.

Toward highly accurate and reliable GNSS positioning

A few years ago, we came across a particularly challenging case. Ultimately, we were granted a patent for how we solved it. The car we were working with was equipped with a wheel tick sensor for speed and z- and x-axis (yaw and roll) gyroscopes that together delivered 2D dead reckoning. When driving up the spiral ramp of a multi-story car park, we observed that the heading – the horizontal direction the car was pointing – determined by dead reckoning was systematically off.

After some investigation, we found out that we could improve the heading by estimating the car’s pitch – the degree to which it pointed up or down. In the end, we figured out a way to do so using the ratio between the yaw and the roll gyroscope readings. As an important aside, the method lets us accurately determine a vehicle’s altitude in the absence of GNSS reception using this reduced set of inertial sensors.

The situation described might sound a bit abstract, and, well, it is. But it is yet another piece in the larger puzzle of fusing the inputs from a variety of sensors into a highly accurate and reliable reading of position, orientation, and speed.

Ultimately, sensor fusion and dead reckoning will contribute to making satellite-based positioning technologies sufficiently robust to meet the stringent demands of highly and fully autonomous driving applications. And if, in the future, you find yourself cursing at your GPS, chances are you’ll have only yourself to blame. 

For more on our dead reckoning solutions, be sure to download our white papers on the subject:

Alexander Somieski

Senior Manager Software Development in the Sensor Fusion Positioning team

You might also be interested in