Hiroki Sawada

RESEARCH & PROFILE

Publications,
Research History
& Profile

I study adaptive human-robot interaction, predictive coding, recurrent neural networks, and embodied artificial intelligence based on the Free Energy Principle.

View PublicationsDownload CV

ICRA 2026

First-author accepted paper

IEEE T-SMC

First-author journal paper

IROS 2024

First-author LBR

FEP / HRI

Core research area

PUBLICATIONS

Selected Publications

[1] 2026

PredVLA: Predictive Sensorimotor Modeling for Sub-Million-Parameter Robot Manipulation

Hiroki Sawada and Shunichi Kasahara

arXiv preprint (arXiv:2608.26673)

Featured

Proposed PredVLA, a compact recurrent controller organized around predicting sensorimotor dynamics and inferring its latent state from prediction error, rather than compressing a conventional observation-to-action policy. With no robot-data pretraining and about 0.68M trainable parameters, it reaches 86.94% mean success on the three short-horizon LIBERO suites (75.35% across all four), against 19.73% for BC-Transformer and 10.26% for BC-LSTM under a matched parameter budget and an identical frozen front end.

PredVLAPredictive CodingRobot ManipulationLIBERO

[2] 2026

Free-Energy-Gated Plasticity for Real-Time Online Motor Learning in Physical Human-Robot Interaction

Hiroki Sawada and Jun Tani

arXiv preprint (arXiv:2608.23000)

Featured

Extended PV-RNN to continuously adapt its synaptic weights and proposed Free-Energy-Gated Plasticity (FEGP), which regulates the effective learning rate according to variational free energy. A randomly initialized network acquired three cyclic motor patterns during real-time physical human-robot interaction without pretraining, replay, or task-boundary signals, showing that the timing of plasticity relative to model-environment mismatch, rather than its average magnitude, preserves previously acquired behaviors.

FEGPPV-RNNOnline LearningTorobo

[3] 2024

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

Featured

Modeled physical human-robot kinaesthetic interaction using a predictive-coding-inspired variational RNN based on the Free Energy Principle. Quantitatively analyzed how the hyperparameter balancing prediction accuracy and model complexity affects human-robot force interaction.

FEPPV-RNNHRITorobo

[5] 2026

CERNet: Class-Embedding Predictive-Coding RNN for Unified Robot Motion, Recognition, and Confidence Estimation

Hiroki Sawada, Alexandre Pitti, and Mathias Quoy

Proceedings in IEEE ICRA 2026 (accepted)

Featured

Proposed CERNet, a hierarchical predictive-coding RNN with class embedding, enabling unified real-time motion generation, recognition, and meta-inference of confidence. Demonstrated the model using the Reachy humanoid robot in both simulation and real-world settings.

CERNetICRA 2026Predictive CodingReachy

RESEARCH HISTORY

Research Experience

Apr. 2026 – present

Body Integration Learning Framework for Highly Skilled Robots (CREST)

Sony Computer Science Laboratories, Humanity Science Project

Project Researcher

Tokyo, Japan

  • Leading the setup of physical human-robot interaction experiments as a Project Researcher.
  • Using the open-source humanoid robot “OpenArm” by Enactic.

Jan. 2025 – Jul. 2025

Proposal and Real-Robot Demonstration of CERNet: A Unified Predictive-Coding RNN for Generation, Recognition, and Confidence Estimation

CY Cergy Paris Université, Équipes Traitement de l’Information et Systèmes Lab

Prof. Alexandre Pitti, Prof. Mathias Quoy

Cergy, France

  • Proposed CERNet, a hierarchical predictive-coding RNN integrating generation, recognition, and confidence estimation into a unified lightweight model.
  • Demonstrated real-time motion generation, intention inference, and confidence meta-inference in a real-world handwriting task.
  • Conducted real-robot experiments using the Reachy humanoid robot by Pollen Robotics.

Sep. 2021 – Jun. 2026

Human-Robot Kinaesthetic Interaction Based on the Free Energy Principle

Okinawa Institute of Science and Technology, Cognitive Neurorobotics Research Unit

Prof. Jun Tani

Okinawa, Japan

  • Modeled physical human-robot interaction using a predictive-coding-inspired variational RNN based on the Free Energy Principle.
  • Explored bidirectional sensorimotor dynamics between humans and robots.
  • Demonstrated the emergence of mutually adaptive motor behaviors through theoretical modeling and incremental learning.

May 2021 – Sep. 2021

Image Recognition Model for Smartphone Robots Using Transfer Learning

Okinawa Institute of Science and Technology, Neural Computation Unit

Prof. Kenji Doya

Okinawa, Japan

  • Built an image recognition model enabling mutual recognition between smartphone robots via transfer learning of a CNN model in TensorFlow.

Jul. 2019 – Jan. 2020 / Jan. 2021 – Apr. 2021

Magnetic Properties of van der Waals Compounds CeTe3 and CeTe2Se

Okinawa Institute of Science and Technology, Quantum Materials Science Unit

Assoc. Prof. Yoshinori Okada

Okinawa, Japan

  • Evaluated magnetic properties of van der Waals layered compounds CeTe3 and CeTe2Se.
  • Published the research results in a peer-reviewed journal.

Apr. 2019 – Mar. 2020

Sample Synthesis for Optimizing Thermoelectric Conversion Efficiency of LaCu1−δS0.5Se0.5O

Keio University, Matoba-Kamihara Laboratory

Prof. Masanori Matoba

Kanagawa, Japan

  • Successfully synthesized polycrystalline LaCu1−δS0.5Se0.5O.
  • Evaluated the effect of Cu defects on thermoelectric properties and published the results.

WORK EXPERIENCE

Engineering Experience

Mar. 2026 – Apr. 2026

Designing Scientific Prompts and Evaluation Frameworks for AI Systems

Mercor (Bilingual Japanese STEM Expert Writer & Editor)

Tokyo, Japan

  • Designed prompt-rubrics for evaluating LLMs in science and engineering.
  • Worked as both a prompt creator (Writer) and an evaluator (Editor).

Sep. 2023 – Mar. 2026

Development and Operation of an Automated Meeting Minutes Delivery System Integrating LLM and NLP

Reazon Holdings Corporation, R&D Division

Tokyo, Japan

  • Employed as a full-time engineer four days per week.
  • Main engineer for an automated meeting minutes delivery system combining LLM and NLP.
  • Built a scalable backend using AWS EC2, S3, SageMaker Asynchronous Inference, and Lambda.
  • Developed a GPT agent using Retrieval-Augmented Generation to answer natural language queries about company regulations.

EDUCATION

Education

Sep. 2020 – Jun. 2026

Okinawa Institute of Science and Technology

Ph.D. in Computer Science (Cognitive Neurorobotics)

Okinawa, Japan

Apr. 2016 – Mar. 2020

Keio University

B.S. in Applied Physics and Physico-Informatics

Kanagawa, Japan

FELLOWSHIPS

Scholarships & Fellowships

Jan. 2025 – Jul. 2025

French Government Scholarship

Government of France

A French government scholarship for Japanese Ph.D. students. Provided 1,709 EUR per month for six months.

Apr. 2021 – Mar. 2022

JEES & SoftBank AI Human Resource Development Scholarship

SoftBank / JEES

An unrestricted 1 million JPY scholarship awarded to selected Ph.D. students in AI.

Sep. 2020 – Jun. 2026

OIST Ph.D. Fellowship

Okinawa Institute of Science and Technology

Selected as an OIST Ph.D. fellow receiving an annual scholarship of 2.4 million JPY.

ACADEMIC SERVICES

Academic Services

May 2025

Reviewer, IEEE ICDL 2025 Contributed Papers

Reviewed full papers in cognitive robotics and learning for the IEEE International Conference on Development and Learning 2025.