How Emesent is using cutting-edge Wildcat SLAM technology to lead the way in autonomous mapping
Emesent’s award winning Hovermap is a drone autonomy and LiDAR mapping payload that uses LiDAR data and advanced algorithms to generate accurate 3D point clouds of the scanned environment, as well as automate the drone flight even when GPS is not available.
Supercharged with a world-class 3D SLAM solution from Australia’s federal research agency
The Robotics and Autonomous Systems Group within CSIRO / Data61 have built a world-wide reputation for pioneering the development of extremely accurate and robust 3D LiDAR-based SLAM solutions since 2008. Wildcat is the latest SLAM implementation from this group, building on more than a decade of experience to provide a new benchmark in accuracy, robustness and processing speed, and is optimized for robotics applications.
CSIRO Data61 and Emesent working together in the latest DARPA SubT Challenge
Strength through ongoing close collaboration
The Emesent Founders and many of the core engineering team were previously part of the Robotics group at CSIRO, working closely with CSIRO SLAM experts to integrate Wildcat with Emesent’s Hovermap drone autonomy and mapping payload. The Founders left CSIRO and formed Emesent to commercialize the Hovermap technology.
The strong connection between Emesent and the CSIRO Robotics and Autonomous Systems Group continues, with ongoing collaboration through the Wildcat Early Adopter Program and joint research projects. Co-location in the Pullenvale high-tech precinct in Brisbane, Australia has also benefited both parties through regular contact and joint testing activities.
Emesent and CSIRO have teamed up to compete in the prestigious DARPA Subterranean Challenge, with Wildcat being a core component for the robots deployed by both parties. This has pushed the development of key Wildcat features such as multi-agent collaborative mapping – essential when a team of robots is working together to explore and map an area. The combined team placed fourth at the SubT Urban Circuit event held in the US earlier this year, and won the prize for most accurately reported object location – testament to the accuracy of the Wildcat solution.
Optimising performance through field testing and know-how
With their background in Robotics, the Emesent R&D team have a deep understanding of the core SLAM algorithms and how to extract the best from these algorithms for autonomy and mapping.
Hovermap has been deployed with customers since early 2017 – first in prototype form through an Early Adopter Program and then commercially since 2019. Over this period it has clocked up thousands of autonomous flight and mapping hours in challenging, diverse environments around the world. This wealth of real-world deployment data has allowed the Emesent team to optimise the SLAM performance by adjusting parameters and providing valuable feedback to the SLAM development team.
SLAM for both mapping and real-time navigation
SLAM uses range and sensor data to estimate the motion of the range sensor itself. Once this motion has been estimated, the range data can be projected into a common coordinate frame to produce a 3D map. The estimated motion can also be used to control the robotic platform carrying the sensors. In the case of Wildcat and Hovermap, Lidar and IMU data are used to create maps and control the drone’s flight.
When Hovermap is in flight on a drone, the Wildcat SLAM algorithm runs in real-time to estimate the position, orientation and speed of the drone with respect to its surroundings, replacing the need for GNSS data. Emesent’s proprietary drone autonomy algorithms build on this SLAM solution to automate the drone flight, allowing it to navigate without GPS, plan safe paths and avoid obstacles. This requires thousands of calculations to be performed every second onboard Hovermap, driving the need for an efficient SLAM solution. The SLAM solution also needs to be extremely robust and perform flawlessly in any environment. A SLAM error could lead to navigation loss – not an acceptable outcome for a drone during flight. Wildcat’s robustness, accuracy and efficiency have proven to be equal to the task.
All the LiDAR data is stored onboard Hovermap and re-processed after the flight using Wildcat in a mode which is optimized for map quality over real-time processing speed. This produces extremely high quality 3D point clouds – arguably the highest quality SLAM-based LiDAR point clouds on the market. The map quality even caught the attention of Velodyne Lidar Inc. who manufacture the Lidar used by Hovermap. Although many companies use their Lidars to build mapping products, they became a Hovermap owner to showcase the quality of the maps that can be produced by the devices. The quality of the point clouds is thanks to the power of Wildcat and Hovermap’s unique design and build quality.
The ability to perform well in real-time for state estimation and offline for map generation is a significant strength of Wildcat over competing SLAM solutions. Others are generally optimized to deliver either accurate mapping or real-time performance.
An ore pass point cloud produced with a Hovermap flight.
Powerful features for rapid, reliable mapping
With Wildcat under the hood, Hovermap offers several unique features which set it apart from the competition:
Mapping and Autonomy: Hovermap is the world’s only plug-and-play device which offers both drone autonomy and SLAM-based mapping. This allows capturing data in otherwise inaccessible areas such as underground mines.
Accurate mapping: Hovermap has proven to produce highly accurate survey-grade point clouds, especially impressive for a SLAM-based mobile mapping system. It typically achieves relative accuracies of +/- 20 mm in general environments, +/- 15 mm in typical underground and indoor environments and +/- 5 mm for close range scanning. Hovermap with Wildcat helped to win the most accurate object detection prize at the DARPA Subterranean Challenge.
Repeatable results: Change detection and deformation monitoring are common applications for customers in a range of industries. This requires producing repeatable results when mapping an area over time. Hovermap has met these requirements and allowed the detection of sub-centimetre changes in the environment.
Care–free motion while mapping: The power of Wildcat makes the system robust to irregular motion, shock and vibration. Other systems require walking or flying very smoothly, and taking care not to make sudden motions which can produce SLAM slips. Hovermap has been mounted to mining vehicles and jolted around while lowered down ore passes, still producing high quality results.
Robust in feature-poor environments: SLAM algorithms rely on the recognition of unique 3D features in the environment. In smooth tunnels or over flat, featureless terrain, some SLAM systems will not perform well. Hovermap has proven to perform even in challenging SLAM-poor environments such as raise bores and tunnels.
Versatile deployment options: Hovermap is not only capable of autonomous drone-based mapping. Its quick release mechanism allows easy deployment as a backpack, vehicle or cable mounted mapping system. Wildcat is able to cope with all these deployment options.
Rapid data processing: Hovermap is able to produce real-time state estimate for drone flight and low resolution point clouds in real-time for online viewing, and can produce high-resolution maps in offline mode at only twice the data capture time.
Start / Stop Scanning while in motion: Hovermap scanning can be started or stopped while in motion. Start a scan while the drone is hovering or while walking when Hovermap is backpack mounted. Other solutions require complex calibration routines or need to be completely motionless whilst starting the scan.
High data capture speed: The robustness of the SLAM algorithm means that Hovermap can be used at higher speeds while capturing data. Speeds up to 5m/s for drone-based capture and 40km/h (11 m/s) for vehicle-based capture are possible.
Support for real-time multi-agent merging and global optimization: Wildcat was built for multi-robot coordinated mapping, so allows the maps from multiple sources to be merged in real-time during capture to produce a unified, globally optimised result.
Multiple per-point attributes: each data point produced by Hovermap includes multiple attributes such intensity, time, range, return number and ring number. These attributes provide valuable filtering possibilities when the point clouds are used to generate derived data outputs.
Emesent is committed to continue building on the success of the Hovermap and Wildcat SLAM combination through ongoing collaboration with the CSIRO team and dedication of its R&D team. Many more improvements and exciting new features are yet to come.