I feel like this has been done many times before, including by the Skydio team when they are at MIT, but DARPA was impressed:

From JHU:

Johns Hopkins APL Helps DARPA OFFSET Program Take Flight

By Amanda Zrebiec

With each second of the video that ticks away, the suspense builds. Joseph Moore launches a fixed-wing unmanned aerial vehicle (UAV) into the air and it’s buffeted by the wind. Undeterred, the UAV goes about its task to navigate around buildings at high speeds in an urban environment.

The wind picks up at points, and the neon-green fixed-wing UAV steadies itself on those occasions. But ultimately, it navigates the course adeptly, coming within about 10 feet of the buildings and steering around them with relative ease. Most importantly: it doesn’t crash.

“That was a gate for us to get through,” said Moore, the project manager of the Research and Exploratory Development Department team that ran the test at Joint Base Lewis-McChord in Washington state this past August. “We’d never tested anything in an actual physical environment, so proving what we did was huge.”

The test was part of the Defense Advanced Research Projects Agency (DARPA) OFFensive Swarm-Enabled Tactics (OFFSET) program, which envisions swarms of up to 250 collaborative autonomous systems providing insights to ground troops as they operate in dense metropolitan environments. The program is about four years old, Moore said, and it’s unique in structure because the two swarm system integrators — Northrop Grumman and Raytheon — are creating the testbeds and simulation environments for crafting tactics for large-scale autonomous swarms in urban environments.

“OFFSET is developing a variety of swarm-enabling technologies,” said Timothy Chung, the DARPA OFFSET program manager, “from a rich repository of swarm tactics, to virtual environments for swarm simulation, to physical testbeds with real robots where these swarm tactics can be demonstrated in real-world settings.”

This specific test was an effort to answer Moore’s team’s central question for this phase of the project, known as sprints: could fixed-wing UAVs have quadcopter UAV agility and mobility but add greater range, endurance and speed, given that they were fixed-wing in form?

“Imagine you have a futuristic sensor on your aircraft that could, theoretically, map the interior of a building and produce a floor plan,” Moore explained. “You want to put that sensor on a fixed-wing UAV, fly really fast and really close to the building, and come away with a rapid interior scan of the floor plan.

“We’re not there yet, but our goal was to control the vehicle at high speeds in an urban, outdoor environment and do multiple passes around the target building without hitting it.”

UAVs are typically thought of as propeller-armed quadcopters, but previous Independent Research and Development (IRAD) work featuring aerobatic maneuvers with fixed-wing UAVs put APL in an advantageous position to push OFFSET’s fourth sprint forward.

The team took that base work from the IRAD, including its Aerobatic Control and Collaboration for Improved Performance in Tactical Evasion and Reconnaissance (ACCPITER) technology, and spent the first six months of the sprint working with virtual aircraft in a virtual world and the final six using a physical aircraft in a virtually constructed environment.

Using a mesh — a virtual map — of previous DARPA field tests in urban environments, the team flew their fixed-wing UAVs in a virtual world on APL’s campus. They tested, developed the proper algorithms and software, and worked to program an essentially “off-the-shelf” aircraft with bespoke APL-developed electronics package and software.

They did all that at the Lab, but until they trekked to Washington in August, they hadn’t tested it in the physical world. The vehicle’s performance in the virtual world was good. The validation in the physical world performance was exceptional.

In fierce winds, the team launched the craft from approximately 100 meters away, successfully completed two passes through the crowded urban area while maintaining three meters of space from the target building, and then pushed the test to a 250-meter launch, flying four times faster than the quadcopters around a bigger building and under an overpass, autonomously and without crashing.

The program’s fifth sprint is underway, and Moore said this period will focus on adding larger numbers of fixed-wing vehicles operating in urban environments together. The groundwork laid in Sprint 4, especially in validating vehicle performance in the physical world, will be crucial as the team moves forward to address more challenging and complex urban swarm scenarios.