Researchers at the U.S. Army Combat Capabilities Development Command's Army Research Laboratory (ARL) and North Carolina State and Oklahoma State universities developed a reinforcement learning technique to enable swarms of unmanned drones to carry out missions while minimizing performance uncertainty.
The Hierarchical Reinforcement Learning (HRL) approach breaks down a global control objective into hierarchies, including multiple small group-level microscopic control and broad swarm-level macroscopic control.
Each hierarchy exhibits a unique learning loop with its own local and global reward functions, and running these loops in parallel significantly compresses learning time.
Compared to centralized reinforcement learning, HRL shortened learning time by the swarm by 80%, and limited loss of optimality (the point at which the difference between benefits and costs is maximized) to 5%.
From U.S. Army Research Laboratory
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found