Daisuke Okanohara

The University of Tokyo, Preferred Networks (Japan)

Papers

Total Citations

114

H-Index

About

Daisuke Okanohara is a researcher whose work bridges machine learning and robotics, with a focus on advancing how systems model complex, real-world data and plan safe motions. His key research areas include conditional density estimation, density ratio estimation, and learning-based collision-free planning. Okanohara made significant contributions to non-parametric statistics with his work on least-squares conditional density estimation, a method that goes beyond simple regression to capture multi-modal, asymmetric, or heteroscedastic distributions—offering a richer understanding of input-output relationships. His two most-cited papers on this topic, published in 2010, have together garnered over 100 citations, underscoring their influence in the field. More recently, Okanohara has applied generative models to robotics, proposing a novel approach using conditional generative adversarial networks (cGANs) to learn collision-free latent spaces for robot motion planning. This work enables planners to optimize arbitrary criteria while guaranteeing safety, representing a notable step toward more flexible and robust autonomous systems. His research continues to impact both theoretical machine learning and practical robotic applications.