Scientists are building AI to control swarms of up to 250 drones using data from how people play video games like StarCraft and Company of Heroes
- University of Buffalo scientists will develop an AI to pilot large drone swarms
- The AI will be based on data collected from video game players
- The team will use game data about tactical decisions as well as biometric data
- Players used in the project will play StarCraft, Company of Heroes, and Stellaris
Researchers at the University of Buffalo have received a $316,000 grant from the Defense Advanced Research Projects Agency (DARPA), an agency funded by the US Department of Defense, to develop an artificial intelligence capable of controlling swarms of up to 250 drones.
To created the experimental AI, scientists from the university’s Artificial Intelligence Institute will study video game players as they pilot autonomous swarms of digital military units in real time strategy games like StarCraft, Stellaris, and Company of Heroes.
The team will collect data on how the players react to a wide variety of different tactical challenges as well as watching how they react to unexpected changes in the terrain or terms of battle.
Researchers at the University of Buffalo’s Artificial Intelligence Institute will study the way video game players make choices in real time strategy games like StarCraft and Company of Heroes to develop an AI that can control swarms of up to 250 drones
‘We don’t want the AI system just to mimic human behavior; we want it to form a deeper understanding of what motivates human actions,’ University of Buffalo’s Souma Chowdhury told the school’s news site.
‘That’s what will lead to more advanced AI.’
The team will also collect a range of biometric data from the players, through eyetracking software and electroencephalograms, which monitors brain activity while they play.
The team hopes that by combining game decision data with real time biometric feedback giving hints at the player’s emotional and cognitive could help large groups of drones independently coordinate complex tasks.
These could include scouting an area to create a 3D map, maintaining a security perimeter around certain locations, and directing a group of drones to seek out specific targets in unexplored terrain.
The team will record in-game data about tactical choices players make, as well as collect biometric data about eye movements and brain wave activity
The research will be supported by a $316,000 grant from DARPA, which has recently begun expanding its efforts to build autonomous drone swarm piloting tools
‘The idea is to eventually scale up to 250 aerial and ground robots, working in highly complex situations,’ Chowdhury said.
‘For example, there may be a sudden loss of visibility due to smoke during an emergency. The robots need to be able to effectively communicate and adapt to challenges like that.’
DARPA has been testing its own internal program for controlling large groups of drones, called Offensive Swarm-Enabled Tactics (OFFSET).
In January, DARPA completed tests at a training facility in Hattiesburg, Mississppi, which showed a mix of drones and autonomous ground vehicles working together to create a map of the simulated battlefield while locating specific targets in the test facility, represented by QR codes.
While the drones operated independently, human observers were able to track their movements and record map data as it was collected by the drones’ onboard cameras and lidar sensors.
DARPA plans to conduct similar tests of its drone swarm technology three more times over the next 18 months.
