
Hiroki Sawada
Project Researcher, Sony Computer Science Laboratories
CV|Sony CSL|Google Scholar|GitHub|LinkedIn
I am a researcher in AI and robotics. I received my B.Eng. in Applied Physics and Physico-Informatics from Keio University and my Ph.D. from the Okinawa Institute of Science and Technology (OIST). I am currently a Project Researcher at Sony Computer Science Laboratories (Sony CSL).
My research focuses on the Free Energy Principle, Predictive Coding, Human-Robot Interaction, and Embodied AI — exploring how robots can generate, recognize, and adapt their behavior through predictive models grounded in real-world sensorimotor interaction.
Recent News
- 2026.09.30ICRA 2026 Paper Published on IEEE Xplore (CERNet)
- 2026.09.14arXiv preprint: Can Compound Robustness Be Inferred from Single-Axis Evaluations? (LIBERO-CTRL)
- 2026.08.28IEEE RO-MAN 2026 DIP-HRI Workshop @ Kitakyushu
- 2026.08.27arXiv preprint: Should a Compact Robot Controller Be Organized as Observation-to-Action Mapping at All? (PredVLA)
- 2026.08.24arXiv preprint: Gating Plasticity by Free Energy for Online Motor Learning (FEGP)
- 2026.06.08ICRA 2026 @ Vienna
- 2025.7.18Research at CY Cergy-Paris Université
- 2024.10.14Participated in IROS2024 @Abu Dhabi
Selected Publications
PredVLA: Predictive Sensorimotor Modeling for Sub-Million-Parameter Robot Manipulation
Hiroki Sawada and Shunichi Kasahara
arXiv 2026|arXiv
Beyond Single-Axis Testing: Paired Evaluation of Compound Robustness in Vision-Language-Action Policies
Hiroki Sawada and Shunichi Kasahara
arXiv 2026|arXiv
Free-Energy-Gated Plasticity for Real-Time Online Motor Learning in Physical Human-Robot Interaction
Hiroki Sawada and Jun Tani
arXiv 2026|arXiv
CERNet: Class-Embedding Predictive-Coding RNN for Unified Robot Motion, Recognition, and Confidence Estimation
Hiroki Sawada, Alexandre Pitti, and Mathias Quoy
Human-Robot Kinaesthetic Interaction Based on The Free Energy Principle
Hiroki Sawada, Wataru Ohata, and Jun Tani
IEEE Transactions on Systems, Man and Cybernetics: Systems, 2024|Paper