Can a small team of programmers, working intensely, make significant new progress?
ARLINGTON, Va. – Yesterday was the deadline for proposals to qualify for DARPA’s new Innovation House initiative, a program administered by George Mason University that will explore whether cloistering small teams of software programmers in close quarters and setting them to a task can produce significant breakthroughs that would be useful for government programs. The first topic is imagery analysis.
Specifically, “DARPA aims to show that small teams of highly focused, collaborative developers operating under extremely short deadlines can make breakthroughs in automatically obtaining meaning from photos, videos, geospatial data and other imagery-related data.”
How small? From two to four people. How highly focused? Well, you have to live together at George Mason for eight weeks (in two four-week sessions), and meet at least five days a week. The only real stipulation is that you need to be 18 years old and a citizen of the United States to participate.
Teams were asked to submit proposals for the avenues they’d like to explore, and winning teams will be housed by George Mason, and given $10,000 up front (another $40,000 is possible if all criteria is met at the end of the project). Team members have to be acting as individuals (not representatives of universities or corporations), and they’ve got to bring their own computers. At the end of the eight weeks, whatever they produce will remain their own property, but the government will be granted a Government Purpose License, which means DARPA can hand the technology over to whomever in the U.S. federal government to see what they can do with it.
But is eight weeks enough to get anything done?
“Based on my experience, that’s a fantastic model for developing software,” said Chris Scotton, CEO of ClearEdge3D. He should know. His firm, working with funding that includes a significant National Science Foundation Grant, is working in this very area of software development. In ClearEdge’s case, they’re attempting to automatically extract features from point clouds, and further, extract features even where there are gaps in the point cloud, such as shadows cast by pipes blocking pipes below them.
Nor does he see DARPA’s initiative as creating unfair competitive products. “What they’re doing is actually trying to foster innovation, and it’s really quite good,” he said. What DARPA is exploring is referred to in the software world as “scrum methodology,” he said, where small teams work on specific software goals, like a specific feature or ability that needs to be added to a software package, in short, focused sprints. Generally, there’s a “scrum master,” who rides herd over the developer team and keeps them focused on the light at the end of the tunnel.
“It’s a highly accountable environment,” he said of the method’s advantages, “as opposed to the old way, which was somewhat unfocused and longer term.”
In this case, which goes well beyond the 3D and lidar space, there is a general dissatisfaction with the amount of information that can be extracted from video, photographs, lidar and other imagery in automated fashion. Analyzing all the data brought in by military drones and the like remains incredibly laborious, especially if that data is captured in 2D.
For this reason, Scotton surmises the programmers and projects will likely be focused on bettering pattern recognition techniques, rather than getting into the kind of computational geometry that is the basis of ClearEdge’s software advances. The former attempts to categorize and describe what’s in the image. The latter makes predictions on what’s missing from the images based on the information that actually is available.
“If they can get some of the smartest minds in pattern recognition algorithms in a room for eight weeks,” Scotton said, “that could really lead to something.”
Exactly what, we’ll have to wait to find out. The project wraps up in early November.
