MOLISENS: a modular MObile LIdar SENsor System to exploit the potential of automotive lidar for geoscientific applications

. We propose a newly developed modular MObile LIdar SENsor System (MOLISENS) to enable new applications for automotive light detection and ranging (lidar) sensors independent of a complete vehicle setup. The stand-alone, modular setup supports both monitoring of dynamic processes and mobile mapping applications based on Simultaneous Localization and Mapping (SLAM) algorithms. The main objective of MOLISENS is to exploit newly emerging perception sensor technologies developed for the automotive industry for geoscientific applications. However, MOLISENS can also be used for other appli- 5 cation areas, such as 3D mapping of buildings or vehicle independent data collection for sensor performance assessment and sensor modeling. Compared to Terrestrial Laser Scanners (TLSs), automotive lidar sensors provide advantages in terms of size (in the order of 10 cm), weight (in the order of 1 kg or less), price (typically between 5 , 000 EUR and 10 , 000 EUR), robustness (typical protection class of IP68), frame rates (typically 10 Hz- 20 Hz), and eye safety of class (typically 1). For these reasons, automotive lidar systems can provide a very useful complement to currently used TLS systems that have their strengths in 10 range and accuracy performance. The MOLISENS hardware setup consists of a sensor unit, a data logger, and a battery pack to support stand-alone and mobile applications. The sensor unit includes the automotive lidar Ouster OS1-64 Gen1, a ublox multi-band active Global Navigation Satellite System (GNSS) with the possibility for Real-Time Kinematic (RTK), and a 9-axis Xsens Inertial Measurement Unit (IMU). Special emphasis was put on the robustness of the individual components of MOLISENS to support operations in rough field and adverse weather conditions. The sensor unit has a standard screw for 15 easy mounting on various platforms. The current setup of MOLISENS has a horizontal field of view of 360 °, a vertical field of view with 45 ° opening angle, a range of 120 m, a spatial resolution of a few cm , and a temporal resolution of 10 Hz- 20 Hz. To evaluate the performance of MOLISENS, we present a comparison between the integrated automotive lidar Ouster OS1-64 and the state of the art TLS RIEGL VZ-6000. The mobile mapping application of MOLISENS has been tested under various conditions and results are shown from two surveys in the Lurgrotte cave system in Austria and a glacier cave in Longyearbreen 20 on Svalbard.

To estimate the quality of automotive lidar point clouds, a RIEGL VZ-6000 3-dimensional (3D) ultra long range TLS (Riegl Laser Measurement Systems GmbH, 2020) was used for ground-truth acquisition. A test setup was designed to compare the accuracy of the VZ-6000 to automotive lidar sensors (Hammer, 2021). The MOLISENS system made it possible to record data with an automotive lidar independent of the vehicle platform.
To support mobile mapping applications, MOLISENS includes a DGPS and an IMU for georeferenced positioning and orientation. For registration of subsequently recorded point clouds into a cumulative point cloud, i.e. for creating a 3D map, 60 the Simultaneous Localization and Mapping (SLAM) algorithm (Bȃlas , a et al., 2021;Zhang and Singh, 2017) LIO-SAM (Shan et al., 2020) is used. Lidar based mobile mapping systems have already been tested in various disciplines such as indoor mapping applications (Tucci et al., 2018), urban mapping applications (Moosmann and Stiller, 2011;Zhang and Singh, 2017;Behley and Stachniss, 2018), and for geoscientific surveys (Bosse et al., 2012;Kukko et al., 2012;Wang et al., 2013). A major advantage of MOLISENS compared to previous systems is the modular setup focused on automotive sensors, that allows to wire harness between data logger and sensor unit 0.7 200 Lithium-Ion (Li-ion) battery 1.7 16.7 · 10.0 · 7.9 total weight 5.5 mains adapter. The environment is scanned by the sensor unit and the transmitted sensor data are recorded by the data logger.
The data can be downloaded via a Local Area Network (LAN) interface for further post-processing on a computer. The weight of the whole setup, i.e., sensor unit, data logger, battery pack, and wire harness, is 5.438kg for mobile measurements (Table 1), 90 which require external batteries. Table 1 shows also the dimensions of the modules of MOLISENS.

MOLISENS (MObile LIdar SENsor System)
sensor unit environment data logger post processing battery for power supply

Sensor unit
The sensor unit consists of the OS1-64 Gen1 which is an automotive rotating lidar sensor, a ublox active multi-band GNSS antenna of the ANN-MB series, and the 9-axis Xsens MTi 630 IMU. Between the OS1-64 and the IMU is a space that heat produced by the OS1-64 can be dissipated. The sensor unit has a 0.635cm (0.25inch) thread, which is a standard camera thread, 95 mountable on handles, tripods, or other standardized setups.

Software
The software stack of the data logger is shown in Figure 2. Although the official operating system for RaspberryPi is Raspbian In ROS, a master called roscore controls and registers all nodes running in the system. Each node, which can be defined as an entity that performs a task, can exchange data with other nodes by publishing or subscribing messages through topics. Topics are communication channels, which are defined by a unique name and a specified type of message that is transported.
ROS officially supports C++, Python, and Lisp but other programming languages are also possible through unofficial channels.
The specific software packages used in MOLISENS are: -Data recording package: We developed a ROS package in Python that provides an easy interface to start and stop the data recording as well as a flexible configuration for the specific requirements of the use case. Both, the sampling rate of either 10Hz or 20Hz of the OS1-64 and the number of points in horizontal direction of either 1,024 or 2,048, can be 125 selected by the user just before the measurements by using the red and green buttons on the data logger. The IMU data are recorded with 200Hz and the GPS data with 1Hz. The ROS package records all messages from the specified topics and creates a time synchronous rosbag file including the data from all sensors. Later, this rosbag file can be used as input to a SLAM algorithm to generate a 3D map of the measurement area.
-Lidar sensor package: The ROS package provided by the sensor manufacturing company was implemented (Ouster Inc., 130 2020a). It provides transforming the raw data from the sensor into point cloud messages and also includes visualization tools to proof that the lidar sensor is correctly mapping the scenario and to check the light intensity of the points. Due to the computational limitations of the RaspberryPi, only raw data are recorded. The recorded raw data are converted into point cloud messages in a post-processing step.
-IMU sensor package: Similar to the lidar sensor, the ROS driver provided by the manufacturer is used (Xsens, 2021).

135
Only the configuration and topic selection was adopted to meet the requirements of our use case. Most of the topics were omitted to increase the performance of the system.
-GPS sensor package: Another self-developed Python package was used to retrieve the National Marine Electronics Association (NMEA) messages from the ublox GPS module. This driver is also able to receive correction data from Radio Technical Commission for Maritime (RTCM) messages through the integrated Networked Transport of RTCM 140 via Internet Protocol (NTRIP) client. By using this NTRIP client, correction data from the external services are included into the GPS module to improve the accuracy of the measurements. The usage of this correction data is called GPS RTK. The precision is below 2.5cm with fixed RTK when the signal from the satellites is clear and also the base station from the correction data service is not far away, i.e., less than 10km, from the GPS module (Dunning, 2018). When the situation is not optimal, the module is still able to reach a precision between 10cm and 45cm with RTK float (Dunning,145 2018).

Power supply
MOLISENS can be powered either with batteries (e.g., Lithium Iron Phosphate (LiFePO4), Li-ion) or with an AC/DC mains adapter that provides a nominal voltage of 24V. The battery supply supports mobile measurements whereas the mains adapter may be used when recorded data are transferred to the post-processing computer. The described setup including the data logger 150 and the sensor unit draws a current of about 1A when data of all three sensors are recorded. We used either a Li-ion-battery with distance between every point in the point cloud and this plane can be calculated. The standard deviation of the distribution of these normal distances, which represent the range errors, was derived ( Figure 3). We investigated two different materials: retroreflective foil and a cardboard with black dull spray paint. Those surfaces need to be in the same plane perpendicular to 185 the observation direction to quantify the deviations of surfaces with high and low reflectivity. This was realized by attaching the materials to a wooden board which was mounted on the wall. We analyzed the reflectance of the materials based on the comparison to Lambertian targets. From a perpendicular angle the reflectivity of the retroreflector was 200% relative to a 100% Lambertian target. The black dull spray paint had a reflectivity of 10% relative to a 100% Lambertian target. These reflectance values represent the ratio of the measured reflectance to the reflectance of a standard 100% Lambertian material (Muckenhuber 190 et al., 2020, Birkebak et al., 2018. The scanned wall in the background was used to model a reference plane. The normal distances between points and reference plane can be calculated by subtracting the thickness of the wooden board and the target thickness. A reflectance threshold was used to select the points representing the target.
We used scanned circles to determine the vertical and horizontal distances for quantifying the angular accuracy. We attached four circles, representing a rectangle, with dimensions of 4.5m · 2m, to a wall. This rectangle was scanned from three different 195 positions which yields six independent vertical and six independent horizontal distances.
The results of the tests on systematic range errors showed that the OS1-64 has a higher standard deviation in the range error distribution compared to a TLS such as the VZ-6000 (Figure 3 (a)). The most significant range errors of up to 25cm occurred when scanning retroreflective targets with the OS1-64 ( Figure 3d). Furthermore, the range errors in this case are not only larger but are also spread out over a large range of values with a standard deviation of 6.9cm. According to the manufacturer, the 200 errors occurring with retroreflectors result from the time walk error. This error is caused by clock errors and is an internal error source of lidar systems. This means that the light returns so strongly that it deforms the shape of the received signal which leads to an error in the estimation of the peak, i.e., the distance measured (Nahler et al., 2020).   (d)). We found that the VZ-6000 shows larger errors when scans are conducted in indoor environments and at low ranges between 5 and 15m. The found performance issues comprise the following artifacts: range errors at highly reflective targets for the OS1-64, areas without data around highly reflective targets for the OS1-64, corner artifacts for the VZ-6000, and multipath artifacts for the VZ-6000. The systematic range errors, which can be considered as noise, can have great impact on mapping micro-  only executed at the very last step when the answer is required by so called "lazy evaluation". The computation is performed on multiple Central Processing Units (CPUs) in parallel. The package is not limited to build-in functions and additional arbitrary functions can be implemented and applied. Furthermore, the package provides tools for visualization, import and export of 230 widely used point cloud formats. Also a direct interface to the powerful open3D and pandas libraries (Zhou et al., 2018; The pandas development team, 2020) is implemented for additional applications. For more details see the documentation on https://virtual-vehicle.github.io/pointcloudset/.

SLAM algorithm
In robotics, SLAM algorithms are a fundamental prerequisite for feedback control, obstacle avoidance, and planning since 235 SLAM allows a robot's six Degrees Of Freedom (DOF) state estimation (Bȃlas , a et al., 2021). Here, we use a SLAM algorithm to generate one cumulative point cloud from a time-series of point clouds. MOLISENS is either mounted on a moving platform or carried along by a person while recording data. The data recording unit uses ROS as middleware and all data is recorded in a rosbag, which includes IMU, lidar, and GNSS data. Each recorded data type in the rosbag has a timestamp. The recorded data is the input for the mapping algorithm LIO-SAM (Shan et al., 2020), which is applied offline in a post-processing step.

240
LIO-SAM uses the lidar odometry data to estimate the six DOF trajectory of the mapping sensor. The state estimation problem is solved by a factor graph. This incorporates IMU-pre-integration, lidar odometry, GNSS data, and loop closure. The system does not depend on continuous GNSS data. Therefore, the GNSS factor is only added when the estimated position covariance is larger than the received GNSS position covariance. The loop closure factor is responsible for detecting whether a new node has a small Euclidean distance to a prior state. If this is detected, the algorithm tries to match the new state to 245 the near, past state. This is especially useful to correct for potential drifts in altitude when GNSS is the only absolute sensor available. These advantages compared to Lidar Odometry and Mapping (LOAM) (Zhang and Singh, 2017) and other previous state of the art algorithms made it well suited for our use cases.

Applications in geoscience
To test the MOLISENS setup in challenging field conditions, two mapping surveys in the Lurgrotte cave system in Austria and 250 in a glacier cave in Longyearbreen on Svalbard have been conducted. The following Section presents the results of these two measurement campaigns.

Application in speleology
The Lurgrotte, a partially water-bearing cave 15km north of Graz in Styria, Austria, was chosen as a study area for MOLISENS.
The approximately 6km long cave passes through the Tanneben massif between the localities of Semriach and Peggau. Parts 255 of the cave are accessible to tourists. The cave is characterized by an abundance of speleothems, water-bearing passages, and a heterogeneous cave geometry in which narrow passages alternate with large chambers, such as the Great Dome. With an area of approximately 5, 100m 2 , the Great Dome is one of the ten largest cave chambers in Austria (Plan and Oberender, 2016). The mapping campaign demonstrated that MOLISENS can provide a cumulative point cloud even without the use of GNSS measurements. Also, the LIO-SAM algorithm was tested on whether it is able to co-register point clouds that were recorded partly outdoor and indoor. More than 300m of complex cave geometry could 265 be scanned with MOLISENS in less than 12 minutes ( Figure 5 (d)). Measurement (1) includes the switch from an outdoor environment to an indoor environment in a single measurement. Measurement (2) was conducted only inside the cave. An overview of the recorded data is given in Table 2. It has to be noted that a final validation of the point cloud data was not possible at this stage. A validation of the data quality and accuracy requires a geodetic reference. A marked closed traverse and local reference points were measured from the cave 270 entrance to the Great Dome after our fieldwork at the Lurgrotte. The drilled mountings of these marks could be used used again for further tests with our system. A valid assessment of the accuracy of the produced map can then be accomplished with this reference.

Application in glaciology
To test MOLISENS for cryospheric applications, a glacier cave was mapped in the glacier Longyearbreen on the Svalbard 275 archipelago, Norway. The morphological changes of glacier caves give information about the englacial water routing. Ice volume changes in caves are common throughout the year and the inter-seasonal comparison of ice dynamics can indicate a change in the hydro-climatic regime of the glacier (Perşoiu and Lauritzen, 2017). Previous work on cold glacier caves in the study area involved geomorphological mapping and seasonal temperature monitoring (Alexander et al., 2020;Guðmundsdóttir, 2011), but detailed 3D measurements of a glacier cave system are typically not available.   kinds of analysis. We recommend to refrain from analyzing features smaller than 10cm in the processed point clouds. The given accuracy of up to 5cm for ranges greater than 50m leads to an increase in noise when the SLAM algorithm is used.
Applying smoothing filters on the cumulative point cloud is recommended. The frequency and magnitude of the distance error when scanning retroreflective surfaces is expected to be reduced with upcoming firmware version. Although, this must be 300 considered if retroreflective targets are used during geo-referencing. Better yet would be to use non-retroreflective target for geo-referencing, e.g. white paper with black markings. The drifts in the SLAM processed point clouds has yet be quantified and workflows for geo referencing have to be tested.
In our fieldwork, MOLISENS has proven to record data in complex environments even without a GNSS signal. With the data from our fieldwork, the mapping algorithm LIO-SAM was able to map the environment and the trajectory of the mapping 305 sensor into a point cloud. This shows that results are possible under the following conditions: snow and ice surfaces arctic weather conditions (−20°C) very narrow spaces (< 1m) rough sensor handling due to rough terrain or narrow spaces

310
The lack of GNSS data for cave measurements caused drifts induced by small propagating errors in the IMU data. These drifts are yet to be quantified.

Other automotive sensors
In addition to lidar, other automotive perception sensors such as radar systems can be integrated into MOLISENS. Modern automotive and traffic monitoring radar sensors typically operate at 24GHz, e.g., Smartmicro TRUGRD Stream (s.m.s, smart 315 microwave sensors GmbH, 2021), or 77GHz (Ramasubramanian and Ramaiah, 2018), e.g., Continental ARS540 (Continental AG, 2016, 2017, have a range up to 300m, and apply Frequency-Modulated Continuous Wave (FMCW) technologies for relative distance and velocity estimation (Patole et al., 2017) and digital beam-forming to control the direction of the emitted wave (Hasch, 2015). In addition to data on object level (i.e., list of detected traffic participants), radar data is typically also provided as radar clusters. Clusters represent radar detections with information like position, velocity, and signal strength.

320
This raw data format allows to develop and apply new algorithms for detecting changes in the backscatter behavior of the environment caused by various geoscientific processes.

Potential applications
We envision MOLISENS as an useful tool in geosciences. The IP level of the OS1-64 allows us to conduct measurements under adverse conditions and rough sensor handling. This opens a wide range of applications ranging from cave mapping, glacier 325 surface analysis to meltwater channel monitoring with the potential to increase our understanding of the drainage systems of glaciers. A 3D model of a glacier cave can be used to parameterize volumetric properties of the cave with the aim of analyzing cave morphology (Gallay et al., 2015;Šupinský et al., 2019). The recorded intensity values for ice surfaces are significantly lower than for surfaces covered with e.g. moraine material or sediments. Hence, the generated data is also a useful basis for change detection of, e.g., the ice surface (Milius and Petters, 2012).

330
The portable nature, low cost, and robustness of MOLISENS opens up for new applications well beyond cave mapping.
Mobile high resolution 3D mapping of glacier fronts using snowmobiles on sea ice is another possible application and could be conducted at relatively high velocities (up to 60-80km/h). Similarly, regular mapping of coastal bluffs susceptible to coastal erosion (e.g. Guégan and Christiansen, 2017), can be undertaken throughout the polar night season that hinder structure-frommotion photogrammetry for large parts of the year in polar regions.

335
Besides mobile measurements, static measurements can be conducted with MOLISENS to record rapid processes in 3D over time with up to 20Hz. The scanner can be placed permanently in an area of interest and in case of an event, the scanning process could be initiated automatically or remotely. Vice versa constant scanning could detect a process happening which would trigger further process chains.
MOLISENS is also a handy teaching tool since it rapidly acquires data at a fraction of the cost of a conventional TLS. In 340 addition, it can be taken along on excursions more easily, and safety concerns are minimal even in large groups due to the laser class 1 rating. Both at the University of Graz and the University Centre in Svalbard (UNIS) it is planned to use MOLISENS for excursions and practicals focusing on cryospheric topics, mapping methods, integrated geological methods, and digital geological techniques (Senger et al., 2021).
Other potential use cases in physical geography are: This list can be further extended since the system can be attached to a wide range of platforms. Tests have been conducted with platforms like cars, agricultural machines, and boats with promising results. Further optimizing weight and power consumption of the system can also enable small Unmanned Aerial Vehicles (UAVs)s as potential platforms. Apart from geoscientific applications, MOLISENS provides an easy to use setup for testing automotive perception sensors for e.g. sensor 365 modeling and sensor Fault Detection, Identification, and Recovery (FDIR) method development.

Conclusion
In this work, we present a newly developed mobile lidar sensor system called MOLISENS. The system combines an automotive lidar with IMU and GNSS. It provides the opportunity to collect 3D data for a wide range of use cases and applications. Besides the hardware we introduced the post-processing tools provided by the two open source packages LIO-SAM and pointcloudset.

370
LIO-SAM is a SLAM algorithm for cumulative point cloud generation, and pointcloudset a Python package for analysis and post-processing of static measurements.
The integration of the automotive lidar OS1-64 and the mobile mapping approach was tested in measurement campaigns in the Lurgrotte cave, Austria and in glacier caves on Longyeabreen, Svalbard. The system offers a flexible, easy to use, and time-efficient way to acquire 3D point cloud, GNSS, and IMU data. The offline SLAM processing resulted in point clouds 375 which can be the basis to investigate numerous geoscientific problems. The robustness of the sensors and the data logger as well as the battery and storage capacities are well suited to demanding fieldwork situations.
In the near future, additional sensors, such as radar and camera, shall be integrated into MOLISENS and further broaden the range of applications. This is possible due to the modular design structure of MOLISENS.