[ad_1]
Finding hikers lost in the forests can be a long and difficult process, as helicopters and drones can not see the thick canopy. Recently, it has been suggested that autonomous drones, capable of sneaking between trees, could facilitate this research. But the GPS signals used to guide the plane may be unreliable or non-existent in the forest environment.
In a paper presented at the conference of the International Symposium on Experimental Robotics next week, MIT researchers describe an autonomous system that allows a fleet of drones to collaboratively search under dense forest vaults. Drones use only on-board computing and wireless communication – no GPS required.
Each autonomous quadrotor drone is equipped with laser rangefinders for estimating position, location and trajectory planning. When the drone flies, it creates an individual 3D map of the terrain. The algorithms help him to recognize unexplored and already searched areas, so he knows when the area is fully mapped. An external ground station merges the individual maps of several drones into a global 3D map that can be monitored by rescuers.
In a real implementation, but not in the current system, drones would be equipped with an object detection to identify a missing hiker. Once located, the drone would mark the hiker's location on the world map. Humans could then use this information to plan a rescue mission.
"We are essentially replacing humans with a fleet of drones to make search more efficient in the search and rescue process," says lead author Yulun Tian, a graduate student in the Department of Aeronautics and Astronautics (AeroAstro). ).
The researchers tested several drones in randomly generated forest simulations, and tested two drones located in a wooded area of the Langley Research Center at NASA. In both experiments, each drone mapped an area of about 20 square meters in about two to five minutes and merged their maps collaboratively in real time. UAVs also performed well on several measures, including the speed and time required to complete the mission, the detection of forest features, and the accurate merging of maps.
The co-authors of the article are: Katherine Liu, PhD student at the MIT Computer and Artificial Intelligence Laboratory (CSAIL) and AeroAstro; Kyel Ok, PhD student at CSAIL and the Department of Electrical and Computer Engineering; Loc Tran and Danette Allen of the NASA Research Center in Langley; Nicholas Roy, AeroAstro professor and researcher at CSAIL; and Jonathan P. How, Professor Richard Cockburn Maclaurin in Aeronautics and Astronautics.
Exploration and cartography
On each drone, the researchers installed a LIDAR system, which creates a 2D scan of surrounding obstacles by launching laser beams and measuring reflected pulses. This can be used to detect trees. However, for drones, individual trees look remarkably similar. If a drone can not recognize a given tree, it can not determine if it has already explored a zone.
The researchers have programmed their drones to identify the orientations of several trees, which is much more distinctive. With this method, when the LIDAR signal returns a group of trees, an algorithm calculates the angles and distances between the trees to identify that group. "Unmanned aerial vehicles can use this signature as a unique signature to indicate if they had ever been to that area or if it was a new area," says Tian.
This feature detection technique helps the ground station to precisely fuse the cards. Drones typically explore a looped area, producing sweeps as they go. The ground station continuously monitors sweeps. When two drones loop to the same group of trees, the ground station merges the maps by calculating the relative transformation between drones, then merging the individual maps to maintain consistent orientations.
"Calculating this relative transformation tells you how to line up the two maps so that they exactly match the appearance of the forest," Tian said.
In the ground station, a robotic navigation software called "simultaneous localization and mapping" (SLAM) – which maps an unknown area and keeps track of an agent inside the area – uses the same. LIDAR input to locate and capture the position of drones. This helps to merge the cards accurately.
The end result is a map with 3D terrain features. Trees appear as blocks of shades of blue to green, depending on the height. The unexplored areas are dark but become gray when mapped by a drone. The integrated trajectory planning software requires a drone to always explore these unexplored dark areas during its flights. Producing a 3D map is more reliable than simply attaching a camera to a drone and monitoring the video stream, says Tian. Video transmission to a central station, for example, requires a lot of bandwidth that may not be available in wooded areas.
More effective search
A key innovation is a new search strategy that allows drones to more effectively explore a domain. According to a more traditional approach, a drone would always look for the unknown area as close as possible. However, this could be in a number of directions from the current position of the drone. The drone usually travels a short distance, then stops to select a new direction.
"It does not respect the dynamics of the drone [movement]Said Tian. "You have to stop and turn, which means it's very inefficient in terms of time and energy, and you can not really pick up speed."
Instead, the researchers' drones explore the area as close as possible while considering their current direction. They believe that it can help drones maintain a more constant speed. This strategy – where the drone tends to spiral – covers a search area much more quickly. "In search and rescue missions, time is very important," Tian said.
In the paper, researchers compared their new research strategy with a traditional method. Compared to this base, researchers' strategy allowed drones to cover many more areas, several minutes faster, and higher average speeds.
A limitation for practical use is that drones must always communicate with an external ground station for card fusion. In their outdoor experience, the researchers had to set up a wireless router connecting each drone to the ground station. In the future, they hope to design drones capable of communicating wirelessly when they approach each other, to merge their cards and to cut off communication when they separate. In this case, the earth station would only be used to monitor the updated global map.
[ad_2]
Source link
