Uploaded March 2025 | Updated September 2026, 3 days ago
Abstract: While specialist robots demonstrate remarkable proficiency in industrial tasks, their reliance on constrained environments limits their deployment in unstructured real-world settings. In contrast, generalist robots aim to perform diverse, free-form tasks in open-ended environments. In this talk, I will present our recent efforts toward enabling such general-purpose robotic capabilities. Specifically, I will discuss (1) developing flexible
control policies with contextual task understanding, (2) leveraging pre-trained vision-language models to enhance open-world generalization, and (3) effectively collecting and utilizing diverse data sources beyond teleoperated demonstrations.
Bio: Fangchen Liu is a fifth-year Ph.D. student at UC Berkeley, advised by Prof. Pieter Abbeel. Prior to that, she obtained an M.S. from UC San Diego, working with Prof. Hao Su, and a B.S. from Peking University. Her current research focuses on developing algorithms and systems for general-purpose embodied agents and robotics, particularly in open-ended perception, generalizable control, and learning beyond teleoperation datasets. In the past, she also spent time at Google, FAIR, and NVIDIA Research, where she worked on reinforcement learning algorithms and their integration with foundation models.
Abstract: While specialist robots demonstrate remarkable proficiency in industrial tasks, their reliance on constrained environments limits their deployment in unstructured real-world settings. In contrast, generalist robots aim to perform diverse, free-form tasks in open-ended environments. In this talk, I will present our recent efforts toward enabling such general-purpose robotic capabilities. Specifically, I will discuss (1) developing flexible
control policies with contextual task understanding, (2) leveraging pre-trained vision-language models to enhance open-world generalization, and (3) effectively collecting and utilizing diverse data sources beyond teleoperated demonstrations.
Bio: Fangchen Liu is a fifth-year Ph.D. student at UC Berkeley, advised by Prof. Pieter Abbeel. Prior to that, she obtained an M.S. from UC San Diego, working with Prof. Hao Su, and a B.S. from Peking University. Her current research focuses on developing algorithms and systems for general-purpose embodied agents and robotics, particularly in open-ended perception, generalizable control, and learning beyond teleoperation datasets. In the past, she also spent time at Google, FAIR, and NVIDIA Research, where she worked on reinforcement learning algorithms and their integration with foundation models.
