|
During natural and man-made disasters, it is critical to have systems in place that can both locate survivors while also navigating and reporting on hazards present in the environment.
Search and rescue operations have successfully used autonomous unmanned aerial vehicle (UAVs) for recovery work, but they rely on existing communications infrastructure to relay real-time data to emergency personnel. However, there are many places and scenarios where communication infrastructure is non-existent, and there is no current way for UAVs to affectively be used without them.
This is exactly the challenge new research by aerospace engineering assistant professor Michael Otte aims to address.
“If you are looking for people in a nuclear disaster like Fukushima, or a similar situation, there are things that are harmful to both the people and the autonomous tools you are sending in to find them. And if you can’t communicate with the autonomous equipment from your ground station, it creates a two-fold problem,” explained Otte. “We can’t communicate, we’re looking for things we want to find, but we’re also concerned that we’ll lose the autonomous vehicles, which means losing any data they’ve collected.”
That lost data could be critical to knowing not only where human targets are located, but where dangerous hazards exist. Otte’s preliminary research demonstrated that with enough UAVs to throw at a problem—where the system is robust enough to survive a partial loss of UAVs—they could learn where those bad things were at the same time they were finding safer routes and locating human targets.
It was during that work that Otte had a eureka moment. “All of a sudden, this really cool understanding happened,” said Otte. “We realized that whether or not your UAV survives, that survival acts like a sensor!”
If the UAV’s course is known, and it doesn’t come back, that information might indicate a hazard in the area it flew. That information can help determine shape of the next UAV’s path, allowing users to alter it in different ways to gather different types of information from the environment.
Otte anticipates that false positives could happen if an UAV just malfunctions rather than encountering a hazard that took it down, but all of those points of data feed into the next set of flights.
“You can then calculate what it the next best path I should send the next UAV on so that I get the most information about what I am looking for in that particular trip, whether it’s about existing hazards, or human survivors,” said Otte. “We are particularly interested in applying to this idea to environments where the hazards may be changing over time. If the hazard is not moving, or is not moving quickly, we might be able to account for those changes in our model.”
Beyond leveraging the UAV’s survival as a way to determine where hazards might exist in the environment, Otte is also addressing how to minimize the loss of other data collected by the vehicle if it fails to return.
While the UAVs can’t send data back to a central base in real time, if they happen to encounter another UAV while out in the field, they can perform short-range information transmissions to one another. So even if one, or more, of the vehicles are lost to a hazard, its data could make it back to base.
The team’s long-term goal is to develop a functional algorithmic system that can be applied to a wide variety of applications—not just search and rescue—where autonomous vehicles are locating targets under possibly adverse conditions without communications infrastructure.
“There are a number of applications in which you only get a yes/no answer if an event occurred somewhere along the UAV’s path,” explained Otte. “When you get a 'yes' you don’t know where on the path it occurred, and if we want to find the location of the thing triggering the yesses, then you have to solve essentially the same problem.”
Otte is curious to explore how these methods could be also be applied to ecological studies such as tracking down rare species of plants or insects based on collected samples—the yesses along the path—that need to be analyzed off-line in a lab.
Otte’s research is funded through a three-year, $590K grant from the Office of Naval Research in partnership with UMD adjunct faculty member Don Sofge, roboticist at the Navy Research Laboratory.
Related Articles:
Diving Deeper into Competition, and Recruitment
UMD Team Wins Inaugural NIST UAS 3.1: FastFind Challenge
Two Clark School teams take top spots in VFS micro air vehicle competition
UMD Student Team Lauded for Award-Winning Drone
CareDx Acquires UMD-linked Transplant Tech Firm
“Gambit” Pays Off in UMD Team’s Search-and-Rescue Competition Win
New algorithms for multi-robot systems in low communication situations
ArtIAMAS receives third-year funding of up to $15.1M
Underwater Robot Competition Makes a Splash
Past and Present: Jacob Moschler
September 22, 2020
|
