RTK and Sensor Fusion for Autonomous Vehicle Navigation & Guidance

As autonomous driving technology continues to improve, the possibility of widespread personal and commercial use of self-driving cars looks more and more realistic. Even now, new improvements in advanced driver assistance systems (ADAS) make it possible to let computers “take the wheel,” at least for short intervals.

For this transformation to continue, the sensors and systems powering these vehicles must become ever more accurate and precise. Yet, many of them put too little emphasis on the one type of technology that can provide an absolute reference point — satellite systems. Perhaps understandably, designers focus on the more immediate data provided by relative positioning sensors like radar, LiDAR, and cameras, despite their limitations.

Figure 1. Various atmospheric disturbances and obstacles can cause GNSS signal delays, creating errors in receiver positioning estimates. (Image: Point One Navigation)

While it may not be obvious, global navigation satellite systems (GNSS) have an enormous upside for autonomous vehicles (AVs). The only problem is that GNSS positioning estimates often aren’t accurate enough for the precision demands of AV and ADAS technologies.

At least, that’s the case without real-time kinematic (RTK) corrections. As automation increases, this simple but powerful tool for refining GNSS positioning can help GNSS provide a much-needed precision boost to the self-driving stack.

The Limits of Most Self-Driving Sensors

In general, AVs and ADAS rely on data from a combination of local sensors to provide information about the vehicle’s relative position in relation to lane markers and objects in the surrounding environment.

These data sources include cameras, radar sensors, LiDAR sensors, and ultrasonic sensors. Each of these inputs provides a different view of the surroundings, and sensor fusion merges these to create a more accurate, comprehensive picture. Nonetheless, each type of sensor has its limits, from weather interference with cameras to noise disturbances with ultrasonic sensors.

Figure 2. RTK corrections use a base station with a known fixed position to cancel out any sources of error in GNSS positioning estimates. (Image: Point One Navigation)

This picture is commonly combined with high-definition (HD) maps, which provide something closer to an absolute reference point for the vehicle’s position. However, maintaining and updating HD maps gets expensive, and they may be out of date as a result.

More recently, vehicle-to-everything (V2X) technology, which allows vehicle sensors to communicate with other cars, traffic infrastructure, and more, has shown promise in helping to fill in the gaps in these data sources. As of yet, though, V2X faces many hurdles to successful implementation and regulatory compliance.

How Autonomy Drives the Need for GNSS

As autonomy increases, so does the importance of accurate positioning — both relative and absolute — for self-driving cars or ADAS. And while the existing sensor stack can provide relatively detailed relative positioning information, relying on HD maps for absolute positioning is often cost-prohibitive.

It’s precisely here that GNSS receivers can be invaluable. GNSS data has already proven valuable in numerous applications, such as:

Navigation and route planning: GNSS has long been used for navigation with human drivers. It now helps with route optimization and changes based on construction or accidents, providing turn-by-turn navigation plans for ADAS.
Localization: Especially at higher levels of localization, GNSS signals can provide critical localization information to clarify the vehicle’s position in relation to its surroundings or a planned route.
Geofencing: GNSS receivers enable geofencing, which can provide an understanding of geographic boundaries for AVs and ADAS. This can be used to define restricted areas or zones for specific driving behaviors, such as lower speed limits for school zones. Fleet managers may also rely on geofencing to monitor and manage vehicle locations.

Figure 3. Achieving centimeter accuracy with RTK requires a dense network of base stations. Point One’s Polaris network features over 1,700 stations across the globe. (Image: Point One Navigation)

Additionally, satellite signals can provide an extra layer of safety by supplementing the data from relative sensors to verify or improve accuracy. These types of redundancy are critical for autonomous vehicles, which still must clear numerous hurdles to achieve wide-scale use and acceptance.

Fine-Tuning GNSS Positioning With RTK

When a GNSS receiver has an open sky with access to signals from at least four different satellites, it can achieve relatively accurate positioning estimates of within a few meters. Yet, obstacles like buildings, trees, or canyons often interfere with signals. In addition, many other types of atmospheric disturbances or technical issues can also throw off positioning estimates.

Whatever the degree of error, the problem is clear: GNSS adds value as a source of absolute positioning, but on its own, its not accurate enough for autonomous vehicle applications.

Fortunately, vehicle manufacturers don’t have to rely on GNSS receivers alone. The cameras and sensors in the AV navigation stack provide critical relative positioning data to help fill in where GNSS is unavailable. But that still doesn’t help provide a more accurate source of absolute positioning. For that, you need RTK, which can refine GNSS positioning estimates to be accurate within a centimeter rather than meters.

The methodology of RTK is simple — by checking GNSS receiver data against a base station with a known, fixed location, the system can cancel out most of the disturbances and delays that cause errors in positioning estimates. Because the location of the base station is known, and it’s experiencing the same GNSS delays as the receiver, the errors drop out to leave precise positioning data. As long as the receiver is within 50 kilometers of a base station, it’s close enough to refine positioning estimates to within a few centimeters. Less than 10 kilometers, and it delivers centimeter accuracy.

Of course, using this method at scale requires an extensive network of RTK base stations to ensure the vehicle is never far from a reliable source of corrections. The more base stations there are, the more accurate the results. Point One’s Polaris network, for instance, features more than 1,700 base stations across the globe. There is already ample coverage across the U.S., Canada, Europe, Australia, and South Korea, and Polaris continues to add more stations every month.

Integrating RTK With the AV Sensor Stack

Figure 4. When satellite signals are unavailable, INS can still rely on a built-in IMU for accurate positioning and guidance. (Image: Point One Navigation)

Although RTK GNSS corrections can provide accurate real-time positioning estimates, this technology is not yet widely used for vehicle automation. Perhaps this is due to the known limits of GNSS. After all, even with RTK, a GNSS receiver can’t provide positioning estimates from inside a tunnel, parking structure, or other source of complete signal obstruction.

At least, it can’t without an integrated source of relative positioning data to “fill in the gaps” when GNSS is unavailable. And that’s possible when combining the correctional capabilities of RTK with the directional positioning data from an inertial measurement unit (IMU). The latter uses gyroscopes and accelerometers to determine things like angle, direction, and velocity, allowing the vehicle to estimate its location even without a direct satellite connection.

When paired with an RTK-enabled GNSS receiver to create a complete inertial navigation system (INS), the IMU creates a powerful source of absolute and relative positioning data in one system. RTK provides highly accurate positioning data whenever satellites and base stations are within range, giving the IMU a precise starting point for directional positioning when those other sources are obscured.

These types of INS are readily available and easily added to an AV or ADAS to provide detailed positioning information. Point One’s Atlas INS has already been deployed in self-driving delivery vehicles for Faction and autonomous race cars at this year’s Indy Autonomous Challenge. INS technologies can also be integrated into construction equipment, robotaxis, drones, and a wide range of AVs.

Keeping AVs and ADAS on Course With RTK

This is an exciting time in robotics and vehicle automation, and sensors of all kinds provide critical data to drive this technology forward. But the sensor stack for AVs and ADAS is incomplete without a reliable source of absolute positioning. With the help of RTK, GNSS receivers can provide just that — a centimeter-accurate estimate of the vehicle’s position in the real world. Add IMU data through sensor fusion, and AVs can keep close tabs on their position in real time, even when GNSS signals are blocked or unavailable.

Building a comprehensive sensor stack for vehicle automation is a highly effective way to overcome the obstacles that remain between current technology and a future where we safely hand over more control to self-driving vehicles.

This article was written by Lucas Mckenna, Director, Europe at Point One Navigation. For more information, go here .