OxTS LIO Guide

To produce a generic aiding (GAD) file using OxTS LIO, four files are required:

• OXTS raw navigation data (.RD)
• INS to LiDAR position offset (.LIP)
• INS to LiDAR rotation offset (.LIR)
• LiDAR data (.PCAP)

See the OxTS Georeferencer guides for more on getting the best LiDAR data collection, the exact same applies to LiDAR odometry.

The RD file used will need the following feature codes present at the time it was logged:

• LiDAR Odometry
• GAD 

Contact support@oxts.com to purchase or obtain trial feature codes. 

The LIP and LIR files describe the position and orientation of the LiDAR respectively, and are relative to the INS frame. The LIP is the x, y and z displacement from the INS measurement point to the LiDAR device, and the LIR is the relative yaw, pitch and roll displacement between the INS measurement axes and the LiDAR device axes, both are text files with the format:

x/yaw measurement

y/pitch measurement 

z/roll measurement 

1 or 0 (indicates if data has been calibrated) 

Use the OxTS Georeferencer hardware tab to create these files easily with a visual aid.

To fully calibrate your LIP and LIR we highly recommend performing a boresight calibration which will significantly improve the LiDAR odometry. A guide on doing this can be found here. The accuracy of the LIR is of most importance, and so this step is a high priority. The accuracy of LiDAR velocity updates will depend strongly on the accuracy of these relative orientation measurements.

The "Getting the best data" section in this guide covers the key parts of getting the best out of your LiDAR and INS data which will in turn result in the best LiDAR odometry possible.

The LiDAR data can be in three different formats, either raw packets in .pcap, .pcapng or the OxTS format LCOM. 

The equipment setup for an OxTS LIO data collection is essentially the same as any other georeferencing application. The main difference is that ideally the LiDAR in a LIO setup is flat on the roof and not angled as is common in surveying setups. Use a live viewer (such as PandarView 2) to ensure that the LiDAR is not seeing bonnet or roof of the car as this will obstruct the data, however this can be removed in post-processing.

You must have an INS and a LiDAR rigidly mounted together (such that they do not move relative to each other). This is best done by having both devices on a mounting solution together. 

It is essential that there is some form of time syncing contained in the LiDAR data, either from PTP or NMEA and PPS, depending on the LiDAR you are using. Both of these methods are supported by OxTS units. 

LiDAR data is best recorded as .pcap files typically using Wireshark. See the guides under OxTS Georeferencer for more on collecting LiDAR data. You can also use .lcom which is a capture of the raw LiDAR data that is recorded directly onto the internal storage of an OxTS unit. 

Networking

Most LiDAR setups will use an ethernet switch to route PTP messages from the RT to the LiDAR. LiDAR data puts a lot of stress on any network. Ensure that your switch is gigabit ethernet and can cope with large flows of data. If you are also using real time corrections (NTRIP) then it might be the case that differential age increases (time between subsequent received correction messages) or it might be that corrections are not received for a period of time. A simple solution for this is to set the LiDAR to send corrections to a specific IP address that is set up on your PC. 

Georeferencing

The best way to test the absolute precision of your LiDAR setup is to georeference a pointcloud. This should be done as a first step to prove your setup before LIO data collection. It can also be used to spot problems in the setup down the line. Use a data collection in perfect GNSS to test the LiDAR setup.

Perhaps the simplest and quickest way to evaluate your setup precision is by georeferencing your boresight data collection. With a well calibrated setup your boresight pointcloud should look very clean, the targets should look crisp, the correct size and shape. You can load your boresight PCAP into Georeferencer and process it with the option Reflectivity threshold= 135. This will give you a very small pointcloud with just the targets in a few seconds. 

Another test you can do is travelling down a street, turning around, and returning down the street in the opposite direction. Georeference this pointcloud. (Ensure GNSS is perfect throughout). If your setup is well calibrated then objects should not appear blurry or in doublets. Poorly calibrated setups will decrease precision when LiDAR data is overlaid from different directions, whereas well-calibrated setups will have the objects overlapping.

As a first step, ensure your setup is calibrated and capable of producing a crisp pointcloud.

Hesai

• [PANDARXT16]
• PANDARXT32
• [XT32M2X]
• [PANDAR40M]
• [PANDAR40P]
• [PANDAR40]
• [PANDAR64]
• [PANDARQT]
• PANDAR128

Ouster

• [OS032]*
• [OS064]*
• [OS0128]*
• [OS116]*
• [OS132]*
• [OS164]*
• [OS1128]*
• [OS232]*
• [OS264]*
• [OS2128]*

Robosense

• [RUBYPLUS]*
• [RUBY]
• [RUBYLITE]
• [HELIOS1615]
• [HELIOS5515]
• [HELIOS16P]
• [LIDARM1]
• [BPEARL]

Velodyne

• [PUCKLITE]
• PUCK
• [ULTRAPUCK]
• [PUCKHIRES]
• ALPHAPRIME
• [PUCK32MR]
• [HDL32E]
• [HDL64E]

[*] These models have configurable angular resolution independently of RPM. When using these models, one must specify horizontal_scanpoints as well as the lidar model.

[Implemented but untested]