Convincing Test: Precise Positioning Outperforms GPS

GPS is indispensable for navigation but in the future autonomous driving will require a system that is much more precise. Telekom’s Precise Positioning solution delivered a totally convincing performance.

On their late-summer tour across Europe our experts Pascal Lambert and Alexander Wolle clocked 2,183 kilometers as they put our Precise Positioning system (see Infobox) through its paces in live operation. From Hannover to Barcelona, over mountains and bridges, through tunnels and large cities the tour was designed to show whether the combination of satellite navigation, mobile communications and the Cloud can locate a vehicle with greater precision than conventional GPS (you can read the journey blog here). A grand total of 24 terabytes of data was collected and then evaluated by different departments over the past three months. Here, as promised, is our report of the findings.

In recent decades the automobile industry has made great progress with driver assistance systems. Functions like the Lane Keeping Assistant and automatic emergency braking increase the safety of vehicles in traffic. Early ADAS (Advanced Driver Assistance System) functions relied largely on perception-based sensors such as ultrasound and cameras as parking aids and on relative positioning to recognize safety risks. Absolute positioning by means of GNSS satellite navigation is now integrated into smart traffic systems to improve the effectiveness of these systems and make new use cases possible.

In a few years we will have much more advanced functions in our cars: functions such as immersive navigation (immersive View), Crowdsourced Maps, V2X coordination (Vehicle-to-Everything, see Infobox) and sooner or later totally autonomous Level 5 driving. Perception-based sensors and relative positioning will continue to be used in vehicles. Cooperative Intelligent Transport Systems (C-ITS) require precise absolute positioning, however, in order to ensure redundancy, safety and high availability even in difficult driving conditions.

Accuracy requirements for the localization of autonomous vehicles vary industry and use case. Take, for example, an autonomous tractor harvesting blueberries in a field. A deviation of just a few centimeters could destroy the crop and could cause serious yield losses for the farmer.

Cars and trucks do not require the same degree of accuracy as working in a field. A typical autobahn lane is 3.50 meters and a car on average two meters wide. So in theory accuracy to within a meter is sufficient to ensure that a vehicle keeps to its lane. In contrast to autonomous tractors, however, cars and trucks cover long distances and repeatedly encounter difficult and unpredictable environments.

That is why older GNSS positioning solutions are unsuitable for today’s automobile industry requirements. Modern vehicles need positioning that is accurate to within centimeters and that just works—always and everywhere. Positioning solutions must deliver:

• Reliable Accuracy: Precision at lane level, fast convergence, uniform coverage and carrier-grade fail safety are indispensable.
• Guaranteed Safety: Compliance with the ASIL standards of safety and integrity is necessary to foster confidence in the autonomous system and to fulfill statutory requirements.
• Flexible Design: Compatibility with standard hardware, vehicle antennas and both older and new computer architectures enables OEMs to use Precise Positioning without affecting budgets or development cycles.

The Californian company Swift Navigation (see Infobox “How Precise Positioning Works”) joined forces with Deutsche Telekom to trial their joint precise positioning solution on a 2,000 km test drive across Europe (click here for the travel blog). The journey through Germany, Switzerland, Italy, France and Spain covered a wide range of driving conditions including dense urban traffic, tree-lined freeways, the Alps and their network of tunnels, and our experts compared the accuracy of positioning with a standard GNSS system with that of positioning corrected by Swift.

The test vehicle was equipped with two parallel GNSS systems: a GNSS antenna suitable for survey purposes as a reference and a dual-band GNSS platform that can show both the corrected and the uncorrected position. Both systems were supported by an inertial measurement device and by wheel odometry for position estimation. The reference antenna established an Internet connection via an LTE router with a Telekom IoT SIM card in order to communicate with the Swift servers.

NB: The following accuracy ratings relate to measurements based on 95 percent of the data (95 Percentile / 2-Sigma).

Over the entire journey the accuracy of the Precise Positioning solution was to within 34 cm (for 95%), compared with the 82 cm achieved by Standard GNSS. Given the accuracy requirement of about one meter in the automobile sector Standard GNSS might appear to be good enough, so why bother with corrections? There are, however, two important reasons to aim for greater accuracy:

1. An accuracy of 95 percent means that positioning is less precise in five percent of cases. As we move toward totally autonomous vehicles we must increase confidence in the accuracy of positions in order to ensure safe operation. An accuracy—or rather inaccuracy—of 82 centimeters does not leave much leeway for correction.

2. Overall accuracy does not say much about how the system works in the most difficult environments. Assisted and autonomous driving must be safe everywhere and not just under unhindered conditions under an open sky.

Let us take a closer look at four environments in which Precise Positioning is really put to the test:

• Urban canyons where signals are blocked by buildings and multi-directional errors such as reflections or scatterings pose a challenge
• Tunnel exits where you have to rely on dead reckoning (see below) until the satellite signal has been restored
• Multi-lane roads on which the vehicle must sure which lane it is in
• Mountains, which frequently affect the satellite signal

In what might also be called ravines of houses satellite signals often rebound from buildings before reaching their recipient, leading to erroneous measurements. Standard GNSS, accurate to within 101 cm, narrowly failed to achieve the accuracy required for advanced use cases in the automobile sector. Precise Positioning by comparison achieved an accuracy of 27 cm and thereby fulfilled automobile industry requirements without problems.

Urban areas are, moreover, one of the most safety-critical environments for assisted driving and autonomous vehicles. Dense traffic, narrow lanes, cyclists and pedestrians hamper safe navigation in cities even without taking into account the additional challenges posed by blocked GNSS signals and multi-directional errors. The results of driving tests like this one show how important GNSS corrections are for the safe operation of vehicles in these densely inhabited areas.

The results for tunnel exits were also convincing. When a vehicle enters a tunnel it loses track of the satellites and can no longer determine a meaningful position. That is why the position in your navigation system system or on your smartphone sometimes does not change when you drive through a tunnel. More sophisticated systems use a method known as dead reckoning. The current position of an object is estimated by means of its last known position, the distance covered and the direction taken. The problem is that the error is cumulative and increases the further the object travels in a GNSS-free environment such as a tunnel.

Standard GNSS achieved an accuracy of 391 cm and was therefore a totally unreliable solution. It is importsant to correct this positioning error as soon as a vehicle leaves the tunnel. Precise Positioning can reduce considerably the time it takes to re-establish a FNSS position. Our tests showed that Precise Positioning’s fast reconvergence improved the accuracy from nearly four meters to 86 cm.

For Level 2+ ADAS a high level of precision in navigating on multi-lane roads is required. In situations like the approach to a toll booth or a lane closure the car must know very clearly which lane it is in so as to join the right lane and Ampelstatus zu bestimmen. In these situations Precise Positioning correction im proved accuracy from 111 cm to just 39 cm.

Mountains pose different challenges for a GNSS system—challenges such as limited satellite visibility, local atmospheric disruptions and signal blockages. Cell network coverage is also often limited in mountainous regions because cell signals encounter similar problems. Fortunately, the IoT SIM cards in the test vehicle were able to rely on the extensive Telekom network and at least two roaming partner networks per country. So a consistent mobile network connection was ensured along the entire 355-kilometer section of the route through the valleys and tunnels of the Alps. It also ensured that Precise Positioning corrections were received while crossing borders. The system maintained a high 26 cm level of accuracy that amountsd to a 2.5-fold improvement on the performance of the uncorrected version.

In driving trials undertaken jointly by Deutsche Telekom and Swift Navigation the Precise Positioning solution’s performance was put to the test in different exacting environments over more than 2,000 kilometers in five countries. Navigating in urban canyons and tunnels, on multi-lane roads and in mountainous terrain, the system consistently exceeded the performance of conventional GNSS systems and achieved an impressive level of accuracy. Precise Positioning thereby does justice to the requirements of a rapidly evolving automobile industry for which safety and reliability are of the greatest importance.

We will shortly be presenting the results of the test drive in a detailed study.