In September 2023, OxTS sent a team to San Francisco to map the city and collect LiDAR pointcloud data.
The aim of this trip was to test the performance of the new RT3000 v4 Inertial Navigation System (INS) and the effectiveness of using OxTS LiDAR Inertial Odometry (LIO) in improving navigation data.
The RT3000 v4 is the latest GNSS/INS technology from OxTS. It combines our latest IMU10 technology with survey-grade GNSS receivers to output centimetre-level position accuracy. OxTS LIO, is the latest software innovation from OxTS that uses LiDAR to constrain position drift in urban canyons.
Why San Francisco?
San Francisco is one of the standard test grounds for autonomous vehicle testing. It offers a unique combination of wide suburban streets, extreme inclines, urban canyons, tunnels, and rural hill passes making it an excellent test site for navigation technology that traditionally relies on GNSS.
The high-rise buildings covering every side of the road and numerous overhead obstacles made for a very challenging environment in which to collect accurate and reliable navigation data. By utilising OxTS LIO the team was able to stabilise and improve the navigation data even in GNSS-denied environments.
To navigate a built up area accurately, or create a survey-grade LiDAR pointcloud in an urban canyon, you need the most precise navigation data possible throughout the entire trajectory. This can be a real challenge when there are a number of GNSS difficulties such as tall buildings, reflective surfaces, or overhead objects. Consequently, navigating a city using GNSS only at centimetre-level accuracy is almost impossible.
INS devices use an IMU to stabilise the computed trajectory, but in particularly difficult urban environments, where GNSS is constantly interrupted, other aiding sources can be utilised to improve performance further. The addition of LiDAR and OxTS LIO enables survey grade navigation data in the toughest environments. The software uses zero velocity updates from a LiDAR sensor to constrain position drift in the absence of GNSS.
Hardware and Software Set-Up
During the test, the team used following equipment: An OxTS INS device (RT3000 v4) and a LiDAR (Hesai XT32). Nothing else is special about the physical setup of equipment to use LIO, the setup is simply the same as any other INS/LiDAR configuration. The OxTS LIO algorithm optimises the navigation data and this runs in post-processing.
Data Improvement
During the test, the team witnessed a significant navigation performance improvement in a number of traditionally difficult environments. This can be seen in the following examples:
Stonestown multi-storey car park
No GNSS is available inside a multistorey car park. Traditionally this would mean that they are ‘off-limits’ to high precision mobile mapping and navigation-related projects. This is because within less than a minute the navigation device trajectory will have a position error of some metres rendering the data useless. In most cases expensive, time consuming static mapping would have to be employed instead.
However, using OxTS LIO, alongside the OxTS RT3000 v4 the area becomes simply another part of the project with no special measures taken. As can be seen below, the RT3000 v4 makes a good effort with the trajectory and OxTS LIO constrains position drift even further leading to an improved navigation output.
As well as improving navigation performance, the RT3000 v4, with OxTS LIO, will also improve pointcloud accuracy. Instead of producing a pointcloud replete with blurring and double vision, the resulting pointcloud is crisp and clean like any other part of the survey due to the improvement in navigation data.
Mission Street
One of the biggest challenges when collecting accurate navigation data in an urban environment is precision and consistency in a global frame. San Francisco’s downtown area features all possible challenges including tunnels, high rise buildings, reflective surfaces, narrow streets, extreme gradients and overhead objects. The constant interruption to GNSS makes it a robust test for autonomous driving applications and similarly for mobile mapping.
In this scenario, it is not just intermittent GNSS updates and IMU drift that cause position error, the reflective surfaces pose a significant risk of multi-path error in which satellite signals have reflected off of buildings causing large position errors. This could make the INS trajectory skittish and unpredictable.
During testing the RT3000 v4 by itself was very accurate, however when coupled with OxTS LIO it was almost seamless. OxTS LIO keeps the INS on a stable and consistent path through the city allowing it to filter out all multi-path signals.
The result of this is a pointcloud that is consistently centimetre precise without any areas that need to be resurveyed:
Washington Street Tunnel
Like with multi-storey car parks, tunnels block access to GNSS signals. This means that traditionally tunnels are surveyed precisely but only relatively and it is accepted that the global coordinate position of the surveyed points will differ greatly from reality but there is little way to verify this. With a combination of the RT3000 v4 and OxTS LIO however the trajectory is kept on track to a much greater degree of accuracy improving not only the clarity of the tunnel pointcloud images but also their georeferenced coordinates.
(Red with OxTS LIO, white without. Simulated (real time) processing)
Test Data
The wide streets of San Francisco and other cities like it, can pose alternative challenges for navigation. LiDAR Odometry algorithms can sometimes struggle to give accurate updates in these environments due to a lack of features for the beams of light from the LiDAR to reflect from.
To understand the performance of the RT3000 v4 and OxTS LIO we manually turned off GNSS updates for 60 second intervals and measured the position drift with and without OxTS LIO turned on.
The following results were obtained for GNSS-denied drift over a 60 second period:
| Measurement | Drift (v4 + LIO) | Drift (v4 only) |
|---|---|---|
| Position 3D | 0.263 m | 0.56 m |
| Position 2D | 0.26 m | 0.538 m |
| Heading | 0.099° | 0.132° |
| Velocity 3D | 0.018 m/s | 0.039 m/s |
| Altitude | 0.066 m | 0.234 m |
| Percentage Error | 0.05% | 0.108% |
While GNSS was absent, the vehicle travelled a distance of 500 m. During this time, the RT3000 v4 performed remarkably well, drifting only 0.1% of the distance travelled. This equated to a position drift of just 56 cm during a GNSS-outage of 60 seconds over a distance of 500 m. When you include OxTS LIO, position drift over the same distance is reduced even further to only 26cm – or 0.05% of the total distance travelled without GNSS.
Altitude drift is also reduced from 23cm to just 7cm using OxTS LIO.
The LIO settings used during the test were mostly default and have not been perfectly optimised so we could expect even better results. This indicates that even in very harsh conditions where GNSS drops out completely for a minute, OxTS navigation technology will stay on track allowing high precision navigation and mapping almost anywhere.
Pointcloud examples...
Conclusion
San Francisco’s myriad of challenges to precise navigation are overcome by the accuracy of the RT3000 v4 and OxTS’ LIO algorithm.
During the test, we witnessed that through a combination of OxTS’ latest hardware and software developments, position drift was constrained significantly during a 60 second GNSS outage.
As mentioned earlier, with tighter optimisation and possible combination with other OxTS technology features like gx/ix tight-coupling technology and the Generic Aiding Data (GAD) interface, performance could be improved even further.
We’re on our way to enabling our customers to truly navigate anywhere.
Further Reading...
Land-based LiDAR Surveying Solution Brief
When architecting your own mobile mapping vehicle, possibly the most important aspect to get right is your source of localisation. It must enable you to accurately georeference your sensor data to a physical place on Earth.
This solution brief steps through the aspects that we recommend our customers consider when deciding on the source of localisation for their mobile mapping vehicles.
Read the solution brief to learn how why the RT3000 v4 may be the right option for you.
LiDAR Inertial Odometry Technical Paper
OxTS LiDAR Inertial Odometry (LIO) uses LiDAR data to constrain position drift in urban canyons to both improve navigation performance and enhance pointcloud accuracy.
This technical paper outlines how LIO works, the potential applications and use cases for the technology and details performance evaluations from tests in London and Oxford.
Read the OxTS LIO technical paper today and learn how you can improve urban canyon navigation performance.
RT3000 v4 datasheet
Combining survey-grade GNSS positioning with OxTS’ best ever inertial measurement unit, the RT3000 v4 offers a robust, out-of-the-box navigation solution for uninterrupted position, orientation and motion data in all environments.
Read the datasheet to for more detail on the specification you can expect from the RT3000 v4.
