Max Robotics — Global Service Robot Directory + US Certification

About

Hirotaka Hachiya is a researcher whose work spans machine learning, reinforcement learning, and robotics, with particular emphasis on developing statistically principled and computationally efficient algorithms. He has made significant contributions to conditional density estimation, introducing least-squares frameworks that go beyond traditional regression to capture multi-modal, asymmetric, and heteroscedastic distributions — work that has accumulated over 100 citations across two closely related papers. In reinforcement learning, Hachiya has tackled fundamental challenges in policy search and value function approximation, developing innovative techniques for sample reuse in policy gradient methods and EM-based policy search that reduce the prohibitive sampling costs common in continuous robot control tasks. His work on geodesic Gaussian kernels and manifold-based value function approximation demonstrates a sophisticated understanding of the geometric structure underlying real-world robotic environments, addressing discontinuities that standard Gaussian kernels fail to capture. His research also extends to robustness, exemplified by his least absolute policy iteration method for handling outliers in reward observations. More recently, Hachiya has explored deep generative approaches, applying variational autoencoders to laser-scan compression for robot self-localization. Collectively, his contributions reflect a career dedicated to bridging rigorous statistical methodology with practical robotics applications.