ROBOTICS · LEARNING · CONTROL

Jinchen
Ruan.

Learning to move.
Moving to understand.

I study robot self-modeling
and action-conditioned 3D generation.

PERSONAL FIELD NOTES / 01SCROLL TO EXPLORENEW YORK, US

01 / SELECTED WORK

Ideas, in motion.

Robot self-modeling. Controllable 3D generation.
2026 / RSS + CoRL

TDCR hardware and RGB-D capture · Figure 1

RSS 2026Accepted

Jiong Lin*, Jinchen Ruan*, Hod Lipson

An action-conditioned flow model predicts a tendon-driven continuum robot’s settled 3D shape from motor commands, evaluated in MuJoCo and on real hardware.

Self-modelingFlow matchingContinuum robots
Paper PDFCode

Shape interpolation × kinematic control · Figure 2

CoRL 2026Accepted

Jiong Lin, Jinchen Ruan, Hod Lipson

Two-stage flow matching learns shape and articulation without a predefined kinematic skeleton. Generated point-cloud sequences can be converted into URDFs and actuated in PyBullet.

Articulated generationFlow matchingNeural simulation
Paper PDFCode

* Equal contribution. Full manuscripts are available via Paper PDF.

Jinchen RuanJR / OFF THE CLOCK
A LITTLE CURIOSITY GOES A LONG WAY.

02 / BEHIND THE RESEARCH

Curiosity,
made physical.

I’m a Ph.D. student at Columbia University, advised by Prof. Hod Lipson in the Creative Machines Lab. I received my M.S. in Mechanical Engineering from Columbia in May 2026.

My research focuses on robot self-modeling, action-conditioned generative models, and continuum robots. Previously, I received my B.Eng. in Robotics Engineering from Beijing University of Technology and worked on robotic control and perception in academia and industry.

RESEARCH FOCUS
Robot self-modelingAction-conditioned generationContinuum robotics
The longer story, in my English CV

03 / LET’S CONNECT

Have an idea?
Let’s talk.