Shoichiro Yamaguchi

Preferred Networks (Japan), Vector Institute

Papers

Total Citations

H-Index

About

Shoichiro Yamaguchi is a robotics researcher whose work bridges the gap between machine learning and real-world robotic control. His primary research areas include collision-free motion planning, policy transfer, and generative models for robotics. Yamaguchi’s most notable contribution is his pioneering use of Conditional Generative Adversarial Networks (cGANs) to learn a latent space that encodes only collision-free robot configurations, conditioned on obstacle maps. This approach, detailed in his 2023 paper (7 citations), allows for efficient, optimization-criteria-agnostic planning, enabling robots to generate diverse and safe trajectories without explicit collision checking. He is also the lead author of the MANGA framework (Method Agnostic Neural-policy Generalization and Adaptation), presented in 2019 and 2020 (totaling 5 citations). MANGA tackles the critical challenge of policy transfer across environments with varying dynamics and noise, decoupling policy learning from system identification to enable robust, adaptable robotic control. While his citation counts are still growing, Yamaguchi’s work represents a significant step toward more flexible, generalizable, and safe autonomous systems, making him a promising voice in modern robotics research.