2025—Now
Sony Interactive Entertainment · FTG
Engineer
Facial-expression AI and rigging research for expressive digital humans.
A visual map of my workTrains → robots → food → digital humans
Yusuke Urakami — AI and robotics engineer
AI / Robotics Engineer · Tokyo
I build intelligent systems that understand movement, expression, and the physical world.
Currently working on facial-expression AI and digital humans at Sony Interactive Entertainment's Future Technology Group.
DoorGym / randomized world sampler
Simulation running
A controlled environment for doors with unpredictable opinions.
01 / About
Engineering across boundaries.
I am an AI and robotics engineer with a foundation in mechanical engineering, controls, electronics, and software. I enjoy taking ideas all the way from physical systems to learning algorithms.
Growing up between California and Japan—and later working in New York, Silicon Valley, Toronto, and Tokyo—shaped how I work: as a cross-border bridge-builder who translates between disciplines, teams, and cultures.
02 / Short CV
A career in things that move.
2021—25
Sony AI / Sony Research
Research Engineer
Cooking robotics and visual-evaluation research, including AI for food plating.
2020—21
Roboeye · Toronto
Robotics Engineer
Robotics startup experience connecting research with practical product development.
2018—20
Panasonic Silicon Valley Lab
Robotics Engineer
Robot learning, open-source simulation, and sim-to-real research—including DoorGym.
2012—18
Central Japan Railway
MAGLEV Development Division
Electrical and control engineering, technical communication, and project management.
Education
Columbia · Keio
Mechanical & System Design Engineering
M.S. Mechanical Engineering, Columbia University. B.E. and M.E. System Design Engineering, Keio University.
03 / Selected work
Robots that open unfamiliar doors.
Open source · Robot learning · CoRL
DoorGym
How can a robot learn one useful skill across many doors, handles, appearances, and physical conditions? DoorGym combines domain randomization, vision, and reinforcement learning to train policies in simulation and transfer them to a real robot.
95%best reported simulated door-opening success rate
04 / Beyond the lab
Physical craft in the age of AI.
In 2025 I attended Tokyo Sushi Academy alongside my AI work. It was a deliberate return to physical craft: learning through the hands, earning trust in kitchens, and seeing judgment that cannot be reduced to a single metric.
Side quests: cooking, jazz, pixel art, dream journaling, and small e-paper projects. The sushi loss function was delicious.
Welcome to my field notes
Why I wanted a small place for robotics, AI, craft, and unfinished thoughts.
DoorGym and useful randomness
A short introduction to domain randomization, robot learning, and opening unfamiliar doors.
06 / Say hello
Building something that has to survive the real world?
I enjoy conversations about embodied AI, expressive digital humans, robot learning, and unusually interdisciplinary problems.
Start a conversation
Human messages preferred. Robots may assist with scheduling.