Jason Chen

I am an undergraduate student at the University of Southern California working with Daniel Seita and Gaurav S. Sukhatme. I'm majoring in Computer Science.

My current research focuses on a data-driven approach to improving policy generalization and robustness by generating synthetic data using video and image generation models for deployment in VLA and imitation learning policies.

I am currently also getting an accelerated master's degree in Computer Science at USC.

profile photo

Selected Publications

* indicates equal contribution. † indicates equal advising.

StereoEngine: Scaling Stereo Bimanual Robot Data without Stereo Cameras

Jason Chen, I-Chun (Arthur) Liu, Gaurav S. Sukhatme, Daniel Seita

StereoEngine is a way to generate stereo robot data while given (1) a ground truth monocular view or (2) no ground truth monocular view.

In Submission, 2027

ExStereo: Lifting 2D Vision-Language-Action Models to 3D with Explicit Stereo Representations

I-Chun (Arthur) Liu*, Jason Chen*, Gaurav S. Sukhatme, Daniel Seita

ExStereo lifts pretrained 2D vision-language-action models to 3D by rendering stereo image pairs into an explicit multi-view stereo representation that the policy attends to for more precise robotic manipulation.

In Submission, 2027

BiREPO: Learning Goal-Conditioned Bimanual Nonprehensile Pose Reconfiguration Primitives on Diverse Objects

Yunshuang Li*, Xinyi Yu*, Jason Chen*, Changyang Zhang, Faiz Aladin, Gaurav S. Sukhatme, Daniel Seita

BiREPO learns goal-conditioned bimanual nonprehensile primitives for pushing, rotating, and flipping diverse objects, enabling zero-shot sim-to-real transfer and multi-step object pose reconfiguration.

In Submission, 2027

Paper Project

CRAFT: Video Diffusion for Bimanual Robot Data Generation

Jason Chen, I-Chun (Arthur) Liu, Gaurav S. Sukhatme, Daniel Seita

CRAFT generates diverse, photorealistic robot training data from simulation using video diffusion.

International Conference on Intelligent Robots and Systems (IROS), 2026

Best Paper at WORLDS: World Models and Spatial Intelligence for Physical AI

ROPA: Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation

ROPA: Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation

Jason Chen, I-Chun (Arthur) Liu, Gaurav S. Sukhatme, Daniel Seita

ROPA is a data augmentation method for bimanual imitation learning that generates 3rd person robot poses.

International Conference on Robotics and Automation (ICRA), 2026

D-CODA: Diffusion for Coordinated Dual-Arm Data Augmentation

D-CODA: Diffusion for Coordinated Dual-Arm Data Augmentation

I-Chun (Arthur) Liu, Jason Chen, Gaurav S. Sukhatme, Daniel Seita

D-CODA is a method for offline data augmentation tailored to eye-in-hand bimanual imitation learning.

Conference on Robot Learning (CoRL), 2025





Awards


Services