ROBOT LEARNING & REINFORCEMENT LEARNING

Sichang Su

Ph.D. student The University of Texas at Austin

Biography

I am a Ph.D. student in Aerospace Engineering at The University of Texas at Austin, co-advised by Dr. Thinh Doan and Dr. Ufuk Topcu.

My research interests lie in reinforcement learning and robotics. My recent work focuses on efficiently adapting robot foundation models to out-of-distribution tasks with minimal human supervision. My long-term goal is to enable robots to solve complex, diverse tasks efficiently in the real world.

I received my M.Sc. in Mechanical Engineering from the National University of Singapore, advised by Dr. Guillaume Adrien Sartoretti, and my B.Eng. in Engineering Mechanics from Zhejiang University. I have also worked with Prof. Shaoping Xiao as a visiting scholar at the University of Iowa.

EDUCATION

2025-present

The University of Texas at Austin

PhD in Aerospace Engineering

2023-2025

National University of Singapore

MSc in Mechanical Engineering

2019-2023

Zhejiang University

BEng in Engineering Mechanics

SELECTED WORK

Research Project

From foundation policies
to real-world adaptation.

PARTS2026
Earbud insertion · Autonomous 2×
Real-world robot learning2026

PARTS

From Pretraining to Proficiency: Real-World Subtask RL for Long-Horizon Manipulation with Minimal Human Intervention

Sichang Su, Benjamin Yang, Zhiyun Deng, Boyuan Liang, Yip Fun Yeung, Zelin Wang, and Lingfeng Sun

PARTS concentrates real-world reinforcement learning on the bottleneck subtasks of long-horizon manipulation. Lightweight residual policies adapt a frozen base policy while reusing its reliable behaviors, improving complete-task performance without human action corrections during rollouts.

Project website
ReinFlow2025
ReinFlow architecture: visual features feed a velocity head and a learned noise-injection network to generate robot actions during online reinforcement learning. NeurIPS 2025
NeurIPS 20252025

ReinFlow

ReinFlow: Fine-tuning Flow Matching Policy with Online Reinforcement Learning

Tonghe Zhang, Chao Yu, Sichang Su, and Yu Wang

ReinFlow fine-tunes pretrained flow-matching policies with online reinforcement learning. A learned noise-injection network enables exploration and tractable policy likelihoods during training, while preserving efficient deterministic flow inference after fine-tuning.

Project website Paper Code

RESEARCH IN PRACTICE

Industry Experience

May–August 2026

Boston, United States

AUTEL US Inc.

Research Intern

Mentor: Dr. Lingfeng Sun

  • Developed real-world reinforcement learning post-training for vision-language-action models on YAM and Franka robots for long-horizon manipulation.
  • Built supervised fine-tuning and inference pipelines for UMI data and collected GELLO teleoperation demonstrations. GitHub
  • Developed a VLM-driven subtask annotator for LeRobot datasets with human verification.
  • Developed an end-to-end RECAP (RL with Experience and Corrections via Advantage-conditioned Policies) pipeline on YAM bimanual arms. GitHub

March–July 2025

Beijing, China

D-Robotics

Research Intern

Mentor: Dr. Chao Yu

  • Investigated online reinforcement learning post-training for generative robot policies, including flow-matching models.
  • Built articulated manipulation environments in SAPIEN, collected expert demonstrations, and trained diffusion policies for multi-task manipulation.
  • Implemented LoRA fine-tuning for OpenVLA-OFT, π₀, and π₀-FAST on LIBERO.

GET IN TOUCH

Let’s talk research.

I welcome conversations about robot learning,
research collaborations, and related opportunities.

sichang_su@utexas.edu