Shengwei Li

I'm a Master of Professional Engineering student in electrical engineering at the University of Sydney. I work on robot learning and control. At the moment that means asking when a policy learned from demonstrations should notice that its own execution has gone wrong.

Before Sydney I studied automation at Jiangnan University, where I wrote papers on model-free control for low-altitude vehicles and on metal detection. I'm looking for an MPhil position in robot learning or learning-based control.

Research

When should an imitation-learned policy raise an alarm?

Ongoing · Aug 2026 –

Clean68% on target
Biased from step 3014% on target
Seed 0, 12 s simulated. On the right every commanded position is offset by 40 px from step 30.

A policy learned from demonstrations can go wrong mid-task without knowing it. I'm testing whether signals the robot can actually observe (how unfamiliar its inputs look, and whether its successive plans agree) detect execution faults early enough to matter, without firing on normal runs.

Publications

  1. A Dual-Channel Metal Detection Method Based on the RIME Optimization Algorithm

    Shengwei Li, et al.

    EI-indexed conference, 2025

  2. Trajectory Tracking Control for Low-Altitude Vehicles Based on Model-Free Dynamic Linearization

    Shengwei Li, et al.

    EI-indexed conference, 2025 · Corresponding author

Education

  1. Master of Professional Engineering (Electrical)
    The University of Sydney

    2026 –

  2. B.Eng. in Automation
    Jiangnan University

    2022 – 2026

Experience

  1. LLM application development intern
    Shanghai Zhiji Information Technology

    2025 – 2026

    Built and deployed a retrieval-augmented Q&A system used daily by medical students, and the dialogue and scoring engine of a virtual standardised-patient exam.

  2. IELTS instructor, part-time
    Dongnan English

    2024 –