ROBOT LEARNING · EMBODIED AI · TOKYO

Building robot learning systems that survive contact with the real world.

I am a robotics researcher and engineer based in Tokyo, working with the Matsuo-Iwasawa Laboratory at the University of Tokyo and AIRoA.

My current work spans vision-language-action models, post-training, robot-data quality, and end-to-end deployment on bimanual systems. I am particularly interested in learning methods that use human feedback efficiently and in data and evaluation pipelines that reveal why a policy succeeds or fails on physical hardware.

Previously, I conducted doctoral research as a PhD candidate at KTH Royal Institute of Technology from January 2021 to May 2025, supervised by Iolanda Leite. That work developed preference-based reinforcement-learning methods using language, temporal highlights, and synthesized queries.

Portrait of Simon Holk

Current affiliations

May 2025–present

Matsuo-Iwasawa Laboratory

Robot foundation models, multimodal demonstration quality, and generalizable robot learning.

December 2025–present

AIRoA

Large-model post-training, robot-data curation, and deployment-grounded evaluation for bimanual manipulation.

Recent news

  • 2026Auditing Instruction–Trajectory Mismatches in Multimodal Robot Demonstrations accepted for publication in IEEE Robotics and Automation Letters.
  • 2025Joined AIRoA to work on large vision-language-action policies and real-robot deployment.
  • 2025Released RAGDP, a training-free retrieval method for accelerating diffusion-policy inference.
  • 2025FLoRA appeared at ICRA 2025, extending preference-based RL to sample-efficient style adaptation.
  • 2024POLITE received nominations for Best Conference Paper, Best Student Paper, and Best Human-Robot Interaction Paper at ICRA.