Oscar Puming Jiang

Oscar Puming Jiang

PhD student

National University of Singapore

Research Interests

Reusable Robot Skills
Generative Policies
Inference-Time Guidance
Agentic Robot Systems

If you're an undergrad interested in robotics research, talk to as many people in the field as you can, including me! I don't have the bandwidth to take on new students right now, but I'd love to help however I can. Feel free to email me to arrange a chat.

About

I am a PhD student at the School of Computing, National University of Singapore, advised by Prof. Harold Soh.

Before joining NUS, I completed a B.A. in Engineering and an M.Eng. at the University of Cambridge, followed by an M.Res. at Imperial College London, where I was supervised by Dr. Nicole Salomons. I received First Class Honours / Distinction across my previous degrees.

Research

My research focuses on the reliable reuse of learned robot skills. Every learned skill works only within some boundary of conditions. I study how to expand that boundary and how to know where it lies, so a robot can tell when a skill can be trusted.

Expanding the Skill Boundary

A policy is trained once, but new objectives and constraints keep arriving. How can we widen where it works, both by shaping what it learns and by steering it at deployment?

Understanding the Skill Boundary

A skill's name says little about where it succeeds. How can we predict where a learned policy works, and state it in a form that planners and agents can use to invoke and compose it correctly?

Selected Publications

View All →
Agent Priors-guided Policy Learning: robot task demonstration

Agent Priors-guided Policy Learning

Puming Jiang*, Tianrun Hu*, Haozhe Du, Yibo Li, Zhiwei Xue, Xinhu Li, Harold Soh

arXiv · LEAP Workshop @ CoRL 2026

Structural priors connect learning reusable robot policies with understanding where they work and composing them for new tasks.

Guided Streaming Stochastic Interpolant Policy: method overview

Guided Streaming Stochastic Interpolant Policy

Puming Jiang*, Meiyi Wang*, Kelvin Lin, Ce Hao, Harold Soh

R:SS 2026 · Oral

Inference-time guidance steers streaming robot policies toward changing objectives without retraining, enabling fast and reactive control.

News

2026-4

Our work GSSIP has been accepted to R:SS 2026. I will present the work orally in Sydney! 🎉

2024-8

Starting my PhD at the National University of Singapore