AI Learns to Cut Wing Drag by 38% Using Simple Simulations

Researchers at the University of Washington, RWTH Aachen, and the University of Michigan have shown that AI agents can learn to reduce fluid friction on complex shapes by training on much simpler computer models. Using a platform called HydroGym, the agents applied reinforcement learning to tweak virtual fluid flow over objects, first mastering a flat channel and then tackling a three‑dimensional airplane wing. The AI achieved a 38 % reduction in drag on the wing model without ever simulating that exact geometry. The study suggests that such AI‑driven shortcuts could accelerate the design of more efficient vehicles and other fluid‑laden systems, potentially enabling simulations of scenarios that are currently too computationally expensive.