Many of us will have watched TV or movie dramas where a bunch of luckless travelers get totally lost in a dense forest. They hear the unmistakable sound of a helicopter nearby and start waving frantically. Most often, the rescuers will fail to see them and the life-threatening adventure starts anew. Being unable to spot hikers under a thick canopy is a very real problem though, but fleets of autonomous drones might offer some hope.
Researchers from MIT and NASA's Langley Research Center have been modifying quadrotor drones to allow a fleet to work together to zip through the trees of a thick forest without the need for GPS navigation, talking to each other wirelessly as they fly.
Each drone is equipped with LiDAR to create detailed maps of the area. Noting that individual trees are much of a muchness to drones using this system, the researchers instructed the fleet to look for clusters of trees and register unique patterns, and then tap into algorithms to determine if an area has already been searched or is yet to be searched. Onboard path-planning software tells each drone to head for yet-to-be-searched areas.
The mapping information from each drone can be stitched together using simultaneous localization and mapping software at a ground station for monitoring by human search teams. In this three-dimensional map, the trees are represented by blue and green blocks and unexplored zones are at first dark, but turn gray as they're being mapped.
Numerous drones were put through their paces in both forest simulations and within a wooded area at the Langley Research Center. Though they lacked object detection technology that would allow them to identify lost hikers, this could be added to the system for real-world applications, meaning that a drone could mark the collaborative map with the location of missing individuals so that human rescuers could make their way on foot and save the day.
In order to make the most of the drone's speed, the team developed a search strategy that saw fleet members search in a spiral pattern, covering the search area faster than flying for a time, then turning, flying again and turning, and so on. The researchers report that each drone managed to map out a 20 square meter area in two to five minutes, with the maps being accurately stitched together in real time.
The team was also happy with the fleet's ability to detect forest features and with the time it took to complete set missions, among other metrics, but did recognize the practical limitation of having the drones communicate wirelessly with a ground station. Future tweaks will seek to address such issues.
A paper detailing the project is due to be presented at next week's International Symposium on Experimental Robotics. You can find out more in the video below.
Source: MIT News