About

We make factory robots that generalize

Founded in Tokyo in 2023, RLWRLD builds the manipulation foundation model layer for industrial robots. Our research is open; our engineering is close to the factory floor.

2023 Founded
Tokyo Headquarters
3 Pilot partners
Team

The people building it

A small team of robotics researchers and engineers who have spent years in the gap between lab benchmarks and factory floors.

Jung-hee Ryu
Jung-hee Ryu
CEO and Co-founder

Former research scientist at JST (Japan Science and Technology Agency), Robotics Group. PhD in robot learning from Tokyo Institute of Technology, 2021. Spent four years building manipulation systems for automotive assembly cells before founding RLWRLD in 2023.

Sota Miyamoto
Sota Miyamoto
CTO and Co-founder

Previously senior engineer at an industrial automation company in Osaka where he built real-time robot vision pipelines for high-mix assembly. Co-inventor on three patents in point-cloud processing for grasping. MSc from Kyoto University.

Elena Vasquez
Elena Vasquez
Head of Research

PhD from ETH Zurich, Autonomous Systems Lab, with focus on tactile sensing and contact-rich manipulation. Former postdoc in dexterous manipulation at a university robotics group. Several published papers on sim-to-real transfer and force-torque feedback policies. Joined RLWRLD in 2024 to lead the research program.

Our Story

Started at the teach pendant, not in a lab

In 2022, Jung-hee spent three months on the floor of an automotive stamping facility outside Nagoya watching robotics engineers re-teach grasp programs every time a new connector variant arrived from a supplier. Each re-teach took two to five days of offline teach-in programming, followed by a week of test cycles. The economics made high-mix robotic assembly impractical at any but the highest production volumes.

The problem was not the robots. It was that every grasp was hard-coded. No model existed that could read a part it had never seen and predict how to pick it up reliably.

RLWRLD was incorporated in Bunkyo-ku in 2023. The first year was foundational research: building the training corpus, designing the point-cloud transformer architecture, and building the RMB-2K benchmark so progress could be measured objectively against held-out industrial part categories. The first paper appeared at CoRL in late 2025. The first pilot deployment followed in early 2026.

We are a small team and intend to stay close to the engineering reality of factory floors: concrete floors, metallic parts under variable lighting, noisy environments, and automation engineers who need the system to work on a Monday morning without anyone from the startup on-site to restart it.

We do not build general-purpose humanoid robots or consumer manipulation products. We build one thing: the model layer that lets an industrial robot cell handle parts it has never seen before.

Find Us

Come work on the hard problems

We are hiring robotics engineers, ML researchers, and systems engineers in Tokyo. If manipulation in unstructured environments is the problem you want to work on, reach out.