photo

I am a PhD student in the JSK Lab at the University of Tokyo. My research focuses on developing learning methods for robotic skills (motion planning and manipulation) which explicitly take advantage of feasibility and capability assessment. Highlighting this motivation, my recently posted preprint introduces a method to construct/reuse a database of reusable experiences (analogous to initial solutions in numerical optimization), specifically designed to accelerate online planning across problems in a user-specified domain. The construction process employs classifiers to efficiently cover the problem space. My RA-L article defines a robot’s feasible error region and develops a policy searching method that identifies optimal policy parameters to maximize the region volume.

Alongside research activities, I am interested in writing fast and well modularized software. A notable work includes plainmp: a highly tuned motion planning library for articulated robots written in C++ with Python bindings. It solves moderately complex planning problems (e.g., 8DOF Fetch in front of table) in less than 1ms on a laptop, which is, to my knowledge, significantly faster compared to standard implementations.

Before starting my PhD, I studied aerospace engineering for my B.Eng and M.Eng. Some of my works from that time are included in my google scholar profile.

Contact

Education

  • 2019.4 - Present: Ph.D. student in Mechano-Informatics, The University of Tokyo
  • 2016.4 - 2018.3: M.Eng. in Aerospace Engineering, The University of Tokyo
  • 2012.4 - 2016.3: M.Eng. in Aerospace Engineering, Osaka Prefecture University

Preprints under review

  • H. Ishida, N. Hiraoka, K. Okada and M. Inaba CoverLib: Classifiers-equipped Experience Library by Iterative Problem Distribution Coverage Maximization for Domain-tuned Motion Planning, submitted to IEEE Transactions on Robotics (revised and resubmitted in Nov 2024), arXiv link.

Publications

  • N. Hiraoka, H. Ishida, T. Hiraoka, K. Kojima, K. Okada, M. Inaba: Sampling-based Global Path Planning using Convex Polytope Approximation for Narrow Collision-free Space of Humanoid. International Journal of Humanoid Robotics (2024), Paper link.

  • H. Ishida, K. Okada and M. Inaba, Classifier-Aided Maximization of Feasible-Error-Region for Robust Manipulation Learning, IEEE Robotics and Automation Letters (2021), Paper link.

  • See more in Google Scholar