Thermal soaring, a technique used by birds and gliders to utilize updrafts of hot air, is an appealing model-problem for studying motion control and how it is learned by animals and engineered autonomous systems. This simulation-based deep-RL system reveals learning bottlenecks in thermal soaring, defines a new efficiency metric, compares learned policy to soaring vultures' data, and finds that trained network neurons divide into function clusters that evolve during learning, posing thermal soaring as a rich model-problem for learning motion control.
Publisher
Nature Communications
Published On
Jun 10, 2024
Authors
Yoav Flato, Roi Harel, Aviv Tamar, Ran Nathan, Tsevi Beatus
Tags
thermal soaring
motion control
deep reinforcement learning
learning bottlenecks
policy comparison
function clusters
Related Publications
Explore these studies to deepen your understanding of the subject.
