Q&A with Rod Bryant

By on 5 June, 2019

In developing our feature on the state of play for autonomous vehicles in Australia for issue 99 of Position magazine, we became intrigued by the bleeding edge of absolute positioning technology, and it’s application in the autonomous road transport environment. We sat down with Rod Bryant, Senior Director of Technology in the Positioning Product Centre at u-blox, to get an insider’s perspective on the complexities of highly accurate positioning systems for the road.

Position: Hi Rod, thanks for being with us. Let’s start at the top: what are the key applications for high precision GNSS and IMS positioning in increasingly automated road transport?

RB: GNSS is a potentially critical sensor for use in autonomous road transport because it’s the only sensor available offering absolute position.  All other sensors used in vehicle automation offer relative position only.  Radar and ultrasonics may be used to provide distance to adjacent vehicles and obstacles, LiDAR and vision offer means to identify and estimate distance to landmarks – but only GNSS can place the vehicle directly on the map.

On multi-lane roads in urban areas, though not deep in the ‘urban canyons’, accuracy of decimetres typically, and 1-metre (95 percent of the time) is achieved by u-blox’s lane accurate positioning solution, which involves dual band high precision GNSS coupled with a 6-degree-of-freedom Inertial Measurement Unit (IMU) and wheel sensors.  This is accurate enough for GNSS to be used in conjunction with other sensors for lane keeping.

GNSS is far more critical for identifying the road segment that the vehicle is on.  For example, it can be used to distinguish between a road and an adjacent slipway or ramp, or between two roads stacked vertically.  For this application, other sensors are far less useful and hence the GNSS position is required to be highly reliable.  For this application, the receiver must report not just its location but also an upper limit on the error known as a Protection Level.

Another area in which GNSS is critical is for Vehicle to Vehicle (V2V) communications, being introduced using a special version of WiFi (according to the IEEE802.11P standard) and using cellular communications.  The primary uses of V2V and V2I (vehicle to infrastructure) communications are to provide warnings for oncoming vehicles, of braking ahead and so on.  Many of these use cases are being enabled or enhanced by exchanging vehicle location and speed, thereby allowing closing distance and rate of closing to be calculated within the vehicle to anticipate potential collisions.

Position: Could you describe the main challenges related to integrating these technologies into a safe, fit-for-purpose system, in terms of: Technical shortcomings, and integration with landmark-based positioning systems?

RB: GNSS in urban areas is heavily affected by obscuration caused by buildings, overpasses, trees and so on. This leads to restricted view of the sky which results in poor geometry (not enough angular spread amongst the satellites for good accuracy).  GNSS is even more subject to the effects of reflections (multipath) from buildings, trees, overhead signs, tunnel entrances and so on.  As a result, standard precision (code-phase based) GNSS is typically limited in accuracy to between five and 10 metres, even when used in conjunction with wheel sensors and inertial sensors.  High precision (i.e carrier phase based) GNSS is capable of cm accuracy under good conditions but, more importantly, it is far more immune to multipath effects.

However, carrier phase based positioning is difficult and, in the past, has been prohibitively expensive for mass market applications such as automotive applications.  Mass market GNSS receiver manufacturers like u-blox have the potential to adapt the designs of GNSS chips that are produced in high volume in order to bridge the gap.  For u-blox, this has required a major investment in specialised engineering resources and in-house development in order to be able to offer a complete solution.  For others, the approach appears to be partnering with existing high precision GNSS players.

Furthermore, for carrier phase based positioning to be useful in the urban environment, it must cope with very frequent obscuration of signals.  This means that the convergence time must be very short.  Without going into details, this demands dual band operation and the use of Real-Time Kinematic (RTK) corrections rather than just precise point positioning (PPP) corrections as are typically associated with dual band GNSS receivers.  The conventional way of supplying RTK corrections is far too cumbersome for mass market applications, as it requires two-way communication between the receiver and the server. A broadcast scheme such as SSR-RTK is needed instead, and such services are being pioneered largely driven by this application.

As mentioned above, for some use cases, a high integrity receiver is required (i.e. one that provides protection levels).  A protection level is an upper limit on the positioning error that must be provided for each of the across-track, along-track and vertical directions. The probability that the error will exceed the protection limit must be lower than a specified value called the Integrity Risk.  Typically, this value is somewhere in the range from 1 in 1,000 per hour down to 1 in 10,000,000 per hour.  Not surprisingly, this is only feasible because the Alert Limits or thresholds on the protection limits applied by the system are several metres rather than, say, one metre.

Aviation GNSS receivers have been subject to such requirements for many years.  However, the automotive use case is vastly more complex for the following reasons:

  1. In aviation, the receiver operates in a benign environment with the only sources of multipath being the aircraft body itself.  Hence the primary sources of large measurement errors are malfunctions in the satellites themselves, which are inherently very rare and can be assumed not to occur simultaneously.  In the automotive environment, multipath is the primary source of large and frequent measurement errors.
  2. In order to eliminate the potential for small bias errors in the measurements to build up into large position errors, aviation receivers typically compute position using what is known as a single epoch least squares solution. The previous position is ignored.  In the automotive application, the required accuracy demands a filtered solution and, typically, a Kalman filter is employed.  Hence the probability of small measurement biases cannot be ignored.
  3. Aviation solutions use code-based positioning whereas the automotive application demands a carrier-phase based solution.  This introduces another source of error that is particularly difficult to handle when computing the protection limit and that is ambiguity error.  Carrier phase measurements are inherently ambiguous.  We know the carrier phase but not, initially, the number of carrier wavelengths between the receiver and the satellite.  Luckily, techniques for solving such problems were devised many years ago.  However, such solutions can be subject to failure and the protection limit calculation must take account of the probabilities associated with ambiguity errors.
  4. In addition to the ambiguity errors, high precision GNSS makes use of RTK or SSR-RTK correction services to correct the systematic errors from the satellites and the atmosphere.  These are subject to integrity risk which must also be taken into account.
  5. Aviation receivers also do not make use of additional sensors whereas, for the automotive application, the use of an IMU and wheel ticks is required.  This brings in further error sources that must be accounted for.
  6. Finally, points 1 and 3 above involve highly non-Gaussian error distributions which are far more difficult to analyse than the Gaussian errors that dominate the aviation GNSS application.

The solution for the above six points and more is still being evolved, but inevitably involves the large scale characterisation of errors and the development of adaptive error models.

Another problem to be overcome is the inadequacy of road mapping.  Autonomous driving will require feature-rich maps with decimetre level accuracy.  Such maps are being developed, but at enormous cost.  It seems clear that the maintenance of such maps will not be possible at the required update rate and some sort of crowdsourcing approach may be needed.  Again GNSS-based absolute positioning will be at the core of such a scheme.

At this stage in the evolution of autonomous automotive systems, the OEMs are, for the most part, keeping the sensor fusion to themselves.  The primary reason is that they want to use multiple sensor systems as cross-checks against each other which allows them to distribute their risks amongst the sensors.  If the sensors were more closely integrated it would be far more difficult to assign risk to individual sensors.  The penalty for this independence is reduced performance and therefore, we should expect an evolution towards tighter integration eventually.  Two examples amongst many of how this could lead to improved performance include:

  1. By allowing the GNSS to work closely with the landmark-based positioning solutions, it should be possible to calibrate out the slowly varying biases in the map registration that commonly afflicts road mapping.
  2. Visual odometry (i.e. motion sensing) could be used in conjunction with GNSS and inertials to reduce the effects of inertial drifts.

Are you aware of key challenges associated with landmark-based positioning, vehicle mounted sensors?

One of the key challenges is the need to calibrate these sensors for mounting errors and issues with the sensors themselves.  This is another example of where GNSS is proving essential, as I understand that GNSS-based positioning is typically used for sensor calibration in these systems.

In your view, will highly automated vehicles rely on a mix of absolute- and landmark-based positioning technologies, or is there foreseeable scope for one category to usurp the other?

In order to distribute risk I see a combination of GNSS-based and landmark-based positioning being used together for the foreseeable future.

Are there any specific challenges (or milestones) associated with the Australian environment that are notable in your view?

Like all jurisdictions, the pace of regulatory change is lagging technology development in this area and will likely set the pace.

I see the sheer size of the Australian land mass as something of a problem for our relatively small population.  Luckily, most of our population is concentrated in a small number of major cities.  I see such centres as being provided with accurate commercial GNSS correction services while the regions may not be well enough served to facilitate lane accurate GNSS-based positioning.  Having said that, the government has committed to a world leading multi-GNSS SBAS service that will serve the entire continent via GNSS-compatible signals. That is a step in the right direction, but what is really needed is a true SSR-RTK correction service which I believe is under consideration.

Position: Many thanks for your time and insights, Rod.

Stay up to date by getting stories like this delivered to your mailbox.
Sign up to receive our free weekly Spatial Source newsletter.

You may also like to read:

, , , , , , , ,


Sign up now to stay up to date about all the news from Spatial Source. You will get a newsletter every week with the latest news.

How spatial data saves time and money
MetroMap gives access to super high-resolution and accurate ...
Bennett and Bennett expands service capabilities
Bennett and Bennett selects NavVis VLX mobile scanning and d...
Q&A with FIG President, Rudolf Staiger
Staiger reflects on his four years at the helm of FIG and th...
CASA releases RPAS roadmap for the next decade
The roadmap sets out the Authority’s intended approach to ...
OGC joins new Metaverse Standards Forum
The Forum aims to foster open standards collaboration betwee...