DGIST's Human-Like Forgetting Boosts Robot Performance

DGIST (Daegu Gyeongbuk Institute of Science and Technology) announced that Professor Park Kyung-jun’s team from the Department of Electrical, Electronics, and Computer Engineering achieved significant performance improvements in autonomous robots by incorporating a human-like ‘forgetting’ function. In environments such as logistics centers, autonomous robots navigate by setting their own movement paths. When encountering obstacles like forklifts, they adjust their routes. However, even after the obstacle is removed, the robots continue to take detours, as they remember the previous situation.

To address this issue, the research team mathematically modeled the human tendency to rapidly spread information about specific events and then quickly forget them. They trained the robots using a collective intelligence algorithm. As a result, in logistics center environments, the robots’ task processing capacity increased by 18%, and average travel time decreased by up to 30.1%. The robots not only automatically avoid obstacles in the moment but also forget the previous memory afterward, efficiently navigating back to their original paths.

This technology is expected to be applicable not only to logistics robots but also to swarm drones, autonomous vehicles, and exploration and rescue robots.