IROS 2026: Shaping Wind-Tunnel Airflow via Online Learning

rsasaki0109 · x · 2026-08-25

This paper introduces a sample-efficient online learning algorithm to control multi-fan wind tunnels and generate desired airflow profiles (uniform, Gaussian, parabolic). It combines a simplified physical model with iterative measurement-based learning, adapting system inputs without an accurate mapping. The technique enables the 'Floaty' soaring robot to exploit surrounding airflow for efficient flight.

Related event: IROS 2026: Online Learning Shapes Wind Tunnel Airflow for UAVs(2 posts)→

Original post →

More from Embodied

Embodied channel →