Model-Based Reinforcement Learning for Control of Strongly Disturbed Unsteady Aerodynamic Flows

What is it about?

Reinforcement learning is a tool from machine learning that has the promise of addressing many challenges of control of disturbed fluid flows. However, it is burdened by the need for an extensive and costly training protocol, which makes it difficult to apply to many flows. Here, we show that model-based reinforcement learning can overcome the obstacles of training by working directly with a model of the flow.

Featured Image