A delivery robot can identify elements in its surroundings by comparing sensory observations to a previously compiled map. That allows the robot to know the direction it is moving on its route.
In late 2021, a delivery robot, developed as part of the LMAD project, was piloted in Jätkäsaari to deliver DB Schenker’s parcels to the area’s residents. Twelve routes were saved on the robot, and it navigated them independently. This was made possible by point cloud maps created in advance, i.e. 3D descriptions of the environment formed through laser scanning. The routes enabled a large section of Jätkäsaari to be saved as a huge number of 3D points, surprisingly similar to a three-dimensional urban landscape.
GIM Robotics, the company behind the Jätkäsaari delivery robot’s robotics, takes advantage of point cloud maps in its robotics engineering. The robot is equipped with sensors that it uses to observe its surroundings. Then, the machine compares its observations to existing point clouds, compiled using similar sensors. This process tells the robot where it is.
The map reveals the locations of stationary obstacles to the robot
In order to complete the deliveries, the robot must be able to avoid collisions with moving and stationary objects while simultaneously travelling on a predetermined route to the correct locations. To achieve this, it must understand where it is in relation to its surroundings, such as obstacles or alternative access points. Point cloud maps are one of the ways in which the environment can be communicated to a robot. They consist of point clouds gathered by one or more laser scanners. The map used by the robot in Jätkäsaari was created in advance, and a human being verified that it corresponds with reality.
Point cloud maps describe their environment in staggering detail. The 3D point clouds can be used to study various elements and their characteristics in an environment, including their dimensions and locations. One of the benefits is the ability to study geometrical features, such as curvature, inclines and flatness. This information, in turn, can be used to generate predictions on the available space suitable for the robot to travel through.
Lasers create a map by combining measurements on distances and angles taken on all the surfaces that reflect back the beam. As a result, the roads, pavements, buildings, posts, bodies of water, trees and all other objects in the surrounding world are drawn on the map. Moving objects are filtered away from the point cloud through repeated scanning and by utilising the robot’s movements. As several measurements are taken from the same point, moving pedestrians, cyclists and drivers can be removed from the data before a map is compiled.
Parked cars will remain on a map, but this poses no problem for the robot. It is able to locate itself even if a car leaves after the map’s creation and before the robot takes its measurements, because parked cars are usually surrounded by larger, permanent buildings or landmarks.
Openly shared point cloud maps are creating a digital Finland
Methods used in point cloud maps have been developed since the end of the 1980s, but major leaps forward have been taken particularly in the past few years. These maps have become more widely used in land surveying, geography and construction. However, it is worth noting that point clouds used as reference material in cartography are often more detailed in comparison to those used in robotics.
‘Robotics and the 3D datasets created – almost as by-products – through robotics will make new and intriguing applications possible, as long as we know how to use them appropriately,’ says Research Professor Antero Kukko from the Finnish Geospatial Research Institute of the National Land Survey of Finland.
He has been studying laser scanning with his colleague Harri Kaartinen for over ten years. The two men are amongst the top hyperspectral laser scanning experts in the world.
‘It would be really useful to have an electronic database and trading forum where various operators could combine their datasets and the maps could be freely utilised by all. Moreover, by adding open data from cities, municipalities and government agencies, which would allow us to combat some of the quality challenges connected to datasets, we could slowly begin to build a genuinely digital Finland,’ Kukko continues.
Point cloud maps work in the dark and all weather conditions
In robotics, point cloud maps are more reliable in densely built environments than satellite navigation (e.g. GPS), which requires a clear line of sight to the sky. A method based on a point cloud works well in street canyons, in the dark, under bridges and in stairwells, where a satellite navigation signal may not be available.
A point cloud map helps the robot move more confidently. Without this pre-existing route information the robot would have to rely solely on its sensors, and its geolocation process would not be as accurate and reliable as it is now that the robot is able to compare its own observations with the point cloud map. If the robot did not have a map to assist it and even one of its sensors malfunctioned, it would run into problems. If that happened, it would be important for a human being to be there to approve all of the robot’s decisions, similarly to the use of autonomous vehicles.
The point cloud map used by the robot and the measurements taken during operation may be useful for others as well in the future. The datasets collected from various locations provide huge potential for a variety of new operating models. For example, point clouds allow us to catalogue elements in an urban environment and detect any changes. Going forward, a robot regularly operating in a single area could generate up-to-date data on the state of traffic signs, railings and trees, and the snow conditions on pavements, to name a few things. In other words, while doing its rounds the robot could help a city target its maintenance and repairs to where they are needed.
Robotics Engineer Henri Varjotie from GIM Robotics, Professor Antero Kukko from the Finnish Geospatial Research Institute of the National Land Survey of Finland, and Technical Expert Juho-Pekka Virtanen from Forum Virium Helsinki were interviewed for this article.
Further Information about the LMAD project
The Last Mile Autonomous Delivery LMAD project’s delivery robot experiment was a response to the growing needs of urban logistics. Last-mile deliveries can be challenging in cities, and the experiment was carried out to assess whether robotics could provide some help. The results were positive. The robot delivered more than 100 parcels with no need for manual intervention during pick-ups. The project also resulted in the launch of a start-up company, sharing the same name.