Attempt to develop a commander AI that moves a large amount of robots from gamer's brain waves
by
In the era of winning the world champion in the game, standing on par with the top 0.2% of humankind, and learning artificial intelligence (AI) from humans in game play, `` Data obtained by measuring gamer's brain waves An attempt to build an AI that will be a command tower that gives instructions to a large number of robots by using it for machine learning '' is being carried out with the cooperation of the Defense Advanced Research Projects Agency (DARPA), a technology media Digital Trends Reports.
Brain Training Future Swarm Robot Armies | Digital Trends
https://www.digitaltrends.com/cool-tech/brain-training-future-swarm-robot-armies/
Professor Souma Chowdhury of the University of Buffalo is conducting research on swarm intelligence , which states that 'individuals behave with simple rules to create sophisticated behavior as a swarm.' Professor Chowdhury says, `` It is known that various applications can be executed by using a lot of simpler and cheaper robots instead of using one million dollar (about 110 million yen) robot. '' He clarifies the idea that modern machine learning algorithms can be used to gain the ability for simple and inexpensive robots to function autonomously as a group.
by Justin Morgan
And, `` By observing closely how 'humans make tactical or strategic decisions in real-time strategy games', we can identify functions and events that motivate human thinking and behavior, '' Professor Chowdhury said. We are conducting research to 'observe the gamer's reaction accurately and let AI learn its brain waves and eye movements.'
Professor Chowdhury has created real-time strategy games like Starcraft and Stellaris . This game uses resources to produce units and controls many agents on the screen to accomplish missions. The length of one game is set to about 5 to 10 minutes, and the level and It seems that you can play 6-7 patterns depending on the setting.
The 25 test players who play this game will be measured brain waves and at the same time track their eye movements with an ultra-high-speed camera. Test play itself has not been done at the time of article creation, but according to Professor Chowdhury, learning to build an algorithm would require 10,000 simulations in the past, but reinforcement of learning with human data It seems that 1000 simulations are enough.
'For example, a disaster response site requires a supervisor. Even with 100 rescue teams, the supervisor doesn't tell each team member exactly what to do. With that decision, the supervisor thinks about the overall tactics, and that's what we want to build, 'said Professor Chowdhury.
DARPA has also invested a budget in Professor Chowdhury's research, and Digital Trends, a technology media, suggests that AI learned from gamers' brain waves and eye movements may be applied as military technology.
Related Posts:
