Li-Heng Lin

Research Engineer at Google DeepMind

Mountain View, California, United States

About

Research Engineer at Google DeepMind working on AI and Robotics

Experience

  • Research Engineer at Google DeepMind
    Aug 2024 - Present · 2 yrs

    Foundational Research - Robotics

  • Graduate Researcher at Stanford University
    Sep 2022 - Aug 2024 · 2 yrs

    Stanford Intelligent and Interactive Autonomous Systems Group (ILIAD) - Advisor: Prof. Dorsa Sadigh - Few-shot Imitation Learning by Retrieving Prior Experiences • Retrieved data with similar motion by using optical flow. • Proposed an auxiliary reconstruction loss to encourage diffusion policies better utilizing retrieved data. - Enabling robots to understand human gestures (CoRL 2023) • Created prompt for Large Language Models to reason about gestures. • Implemented the whole system and achieve 70% higher success rates than baseline.

  • National Taiwan University (2 yrs 5 mos)
    • Undergraduate Research Assistant
      Jun 2020 - Jan 2022 · 1 yr 8 mos

      Computational Learning Lab - Advisor: Prof. Hsuan-Tien Lin & Dr. Chun-Liang Li - Practical Guide for Deep Active Learning (DAL) • Investigated the effect of several design choices (model initialization, hyper-parameters tuning) in DAL.

    • Undergraduate Research Assistant
      Sep 2019 - Jan 2022 · 2 yrs 5 mos

      Cyber-Physical Systems Lab - Advisor: Prof. Chung-Wei Lin & Prof. Iris Hui-Ru Jiang - Graph-based Intelligent Intersection Management (IEEE IV 2022) • Ensured deadlock free by proposing a protection mechanism based on limiting the number of vehicles. • Reduced vehicle wait time by 52% on average compared to traditional traffic light systems.

  • Software Engineering Intern at Google
    Jun 2021 - Sep 2021 · 4 mos

    Android Accessibility Team - Host: Richard Chang (richardhchang) - Braille Image Translator • Implemented an Android application that translates an image of a braille device into its corresponding text.