Unmanned aerial vehicle (UAV) systems are used by certain Canadian security services in urban areas. Equipped with cameras, these technologies can reconstruct road accidents or participate in criminal investigations. UAVs are undoubtedly effective for missions in hard-to-reach places. But often, even small autonomous drones cannot carry out their research, for example in dense forests, because the GPS signals used to guide them have a hard time passing through the tree canopy, if they can at all. A team of researchers from the Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology and NASA’s Langley Research Center has designed a system that allows drones to explore canopies without using GPS.
Search for Missing Persons Using a Digital Map
The technology consists of a fleet of quadcopter drones and a system that coordinates their tasks. Each drone uses algorithms and wireless communication to visualize the areas. The quadcopter is a type of UAV that can be equipped with a digital camera, offering optimal ground visualization due to its stability.
During flight, the drone creates a 3D map of the terrain. To do this, it is equipped with a LiDAR 2D (Light Detection and Ranging) system that maps the terrain by projecting a laser beam and measuring the time it takes the beam to return.
The algorithms help the drone recognize unexplored areas and those that have already been covered. Researchers programmed the LiDAR system to calculate angles and distances between trees and identify them de facto as being part of a group belonging to a specific area. This way, the system is able to distinguish areas that have been mapped, using colour coding. A ground station communicates with the drone swarm through a wireless router. It also merges the results of each device to obtain a single 3D map. The station is equipped with a robotic navigation software called Simultaneous Localization and Mapping (SLAM). In addition, it uses LiDAR data to locate and detect drone positions. This technique allows it to combine map renderings accurately and in real time. The final map allows rescuers to search more easily for missing persons. When the system is commercialized, the drones will be equipped with an object detection technology to identify people and instantly indicate their position, on Google Maps for example.
According to Yulun Tian, lead author of the research, generating a 3D map is more reliable than monitoring a video stream recorded by camera. Video transmission to a central station requires a large bandwidth, which is often not available or very limited in forests, slowing down rescue work. In this type of mission, time is of the essence. The drone swarm technology is considered innovative because it optimizes search time by coordinating the work of each device. It prevents them from changing direction by ordering each one to sweep their area in a spiral formation.
Tests and Prospects
The researchers performed two tests. Two drones flew over a forest model and were also tested in an actual outdoor environment. Both drones travelled through a wooded area at the Langley Research Center. In both experiments, each drone took two to five minutes to map a space measuring approximately 20 square metres. Both devices were able to generate and merge their maps in real time. The drones also performed well on a number of fronts, including speed and time required to complete the mission, detecting forest characteristics, and accurately merging maps. In the future, researchers want to design drones capable of communicating with each other without a router, to merge their maps and to cut off communication when they separate. At that point, the station will only be needed to monitor the updating of the final map.
This research was presented at the International Symposium on Experimental Robotics Conference in Buenos Aires, Argentina, November 5–8, 2018.