01 · About

Kaiyuan (Eric) Chen

I am a postdoctoral scholar in Computer Science at UC Berkeley, working with Prof. Ken Goldberg in the Berkeley AUTOLab, part of Berkeley AI Research (BAIR). I received my Ph.D. from UC Berkeley in 2025, co-advised by Prof. John Kubiatowicz and Prof. Ken Goldberg, and my B.S. summa cum laude from UCLA in 2020, where I worked with Prof. Songwu Lu on wireless networking.

I build the data and systems infrastructure that lets robots learn and act at scale: robot data management with ICRA 2025 Best Robot Learning Paper, large-scale robot-data-to-VLM pipelines with Robo2VLM (NeurIPS 2025 Spotlight), and cloud robotics with FogROS2, which ships as part of the official ROS 2 release.

Most recently I co-lead Graph-as-Policy (GaP), a multi-agent, self-learning harness in which LLM agents write, verify, and optimize executable behavior graphs for variational automation and robotics tasks, in collaboration with NVIDIA. Previously, I was one of six core developers of SkyPilot, and spent summers at Bosch Research and VMware Research.

Robot Learning Robot Data & Network Systems Distributed Robot Systems Vision-Language Models
02 · Education

Education

Ph.D. — UC BerkeleyComputer Science, 2020 – 2025
Advisors: John Kubiatowicz & Ken Goldberg
M.S. — UC BerkeleyComputer Science, 2024
B.S. — UCLAComputer Science, 2016 – 2020
Summa cum laude
03 · News

Recent News

  • Jul 2026🚀 Released GaP: Graph-as-Policy, in collaboration with NVIDIA — paper (arXiv:2607.05369), project page, code, and docs are live.
  • 2026RoboSQ, semantic queries for task-aligned robot training data, accepted to ICRA 2026.
  • 2026RoboSegNet, multi-task robot trajectory segmentation from visual and kinematic streams, accepted to CVPR 2026 (Findings).
  • 2026HarborMaster, rollback detection for trusted distributed computing, accepted to VLDB 2026.
  • Sep 2025Robo2VLM accepted as a Spotlight (56 of 1,995) at the NeurIPS 2025 Datasets & Benchmarks track.
  • Sep 2025Gave an invited talk at the Berkeley Humanoid Intelligence Center and was interviewed for IEEE Robotics & Automation Magazine's “Mastering Research” series.
Older news
  • May 2025🏆 Robo-DM won the Best Robot Learning Paper Award at ICRA 2025 (1 of 4,250 submissions), and my mentees won the Undergraduate Research Award at ICRA 2025 Art in Robotics.
  • May 2025🎓 Received my Ph.D. from UC Berkeley and stayed on as a postdoctoral scholar with Prof. Ken Goldberg.
  • Jan 2025Joined SkyPilot (Berkeley Sky Computing Lab) as one of six core developers, through May 2025.
  • Oct 2024FogROS2-FT named Best Paper Finalist and Best Student Paper Finalist at IROS 2024 (4 of 3,344 submissions).
  • 2023–25Won Google–BAIR Commons research proposals three years in a row (top ~10%).
04 · Research Spotlight

Graph-as-Policy

GaP pipeline: LLM agents turn a task description and skill library into a behavior graph that self-learns in simulation and executes on real robots
New · 2026 Preprint

GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness for Variational Automation Tasks

Kaiyuan Chen*, Shuangyu Xie*, Letian Fu, Justin Yu, and 20 co-authors, incl. Yuke Zhu, Linxi "Jim" Fan, and Ken Goldberg  (*equal contribution)

GaP represents a robot policy as an executable behavior graph: a harness of LLM agents authors the graph from a plain-language task description, grounds it in a modular open robot-skills library, statically verifies it, and self-improves it in NVIDIA Isaac simulation before executing on real UR5e and Franka robots. Across grocery fulfillment, packing, popcorn making, cable insertion, and bimanual crate washing, GaP reaches 0.93–0.99 success under pose variation — including 25/25 real-world grocery-fulfillment instances and 18.3 bimanual crate-washing cycles per hour.

In collaboration with NVIDIA.

🌐 Project Page 📄 arXiv </> Code 📚 Docs 🧩 Skills Library
05 · Publications

Selected Publications

* denotes equal contribution · full record on Google Scholar

FogROS2-FT fault-tolerant architecture figure
IROS 2024 Best Paper & Best Student Paper Finalist · 4 of 3,344

FogROS2-FT: Fault Tolerant Cloud Robotics

Kaiyuan Chen, Kush Hari, Trinity Chung, Michael Wang, Nan Tian, Jeffrey Ichnowski, Christian Juette, Ren Liu, John Kubiatowicz, Ion Stoica, Ken Goldberg

arXiv Website

All Publications

Preprints & Under Review

Conference & Journal

Workshop papers & abstracts (9)
  • 2025
    Robo2VLM: Visual Question Answering from Large-Scale In-the-Wild Robot Manipulation Datasets
    Kaiyuan Chen*, Shuangyu Xie*, Zehan Ma, Pannag Sanketi, Ken Goldberg · RSS Workshop on Learned Robot Representations (RoboReps)
  • 2025
    Generative AI and Minimalist SculptureUndergraduate Research Award
    Kavish Kondap, Andy Bass, Elle O. Hill, Paloma Raffle, Andrew Goldberg, Tianshuang Qiu, Zehan Ma, Letian Fu, Justin Kerr, Huang Huang, Kaiyuan Chen, Kuan Fang, Ken Goldberg · ICRA Art in Robotics Special Session
  • 2024
    Robo-DM: Efficient Robot Big Data Management
    Kaiyuan Chen, Letian Fu, Siyuan Fu, Lawrence Yunliang Chen, Huang Huang, Kush Hari, Ashwin Balakrishna, Pannag R. Sanketi, John Kubiatowicz, Ken Goldberg · CoRL Open X-Embodiment Workshop
  • 2024
    Blox-Net: Generative Design-for-Robot-Assembly Using VLM Supervision, Physics Simulation, and a Robot with Reset
    Andrew Goldberg, Kavish Kondap, Tianshuang Qiu, Zehan Ma, Letian Fu, Justin Kerr, Huang Huang, Kaiyuan Chen, Kuan Fang, Ken Goldberg · CoRL Workshop on Learning Robotic Assembly of Industrial and Everyday Objects
  • 2024
    RoboDM: Cloud-Based Large-Scale Data Collection
    Kaiyuan Chen, Kush Hari, Lawrence Yunliang Chen, Ken Goldberg · IROS Workshop on Collecting, Managing and Utilizing Data Through Embodied Robots
  • 2024
    Fog-RT-X: Cloud-Based Large-Scale Collection and Management Framework
    Kaiyuan Chen · ICRA Open X-Embodiment Workshop
  • 2023
    Paranoid Stateful Lambdas
    Alex Thomas, Shubham Mishra, Kaiyuan Chen, John Kubiatowicz · SOSP
  • 2019
    Intelligent Wi-Fi Multicast for NDN AR
    Jinghao Zhao, Kaiyuan Chen, Zengwen Yuan, Songwu Lu · ICN WEN Workshop
  • 2019
    Classification of Large-Scale Lyme Disease Data
    Kaiyuan Chen, Rong Huang, Diyi Liu, Catherine Wahlenmayer, Jiewen Wang, Deanna Needell · Joint Mathematics Meeting (JMM)
06 · Recognition

Honors & Awards

🏆

Best Robot Learning Paper Award, ICRA 2025

As first author · 1 of 4,250 submissions (Robo-DM)

🥇

Best Paper & Best Student Paper Finalist, IROS 2024

As first author · 4 of 3,344 submissions (FogROS2-FT)

✨

Spotlight, NeurIPS 2025 Datasets & Benchmarks

56 of 1,995 submissions (Robo2VLM)

🎨

Undergraduate Student Research Award, ICRA 2025 Art in Robotics

With my undergraduate mentees

💡

Google–BAIR Commons Winner, 2023 · 2024 · 2025

Winning research proposals three consecutive years (top ~10%)

🎓

Summa Cum Laude, UCLA Computer Science

Highest Latin honor (top 1%) · Dean's List

07 · Experience

Experience

  • Postdoctoral Scholar, UC Berkeley AUTOLab / BAIR, with Prof. Ken Goldberg — robot data infrastructure & robot learning systems May 2025 – present
  • Core Developer, SkyPilot One of six core developers of the inter-cloud brokerage platform from the Berkeley Sky Computing Lab Jan – May 2025
  • Research Intern, Bosch Research Reliable cloud-robotics infrastructure on FogROS2 — led to adoptions, patents, and publications · with Dr. Nan Tian & Christian Juette Summers 2023, 2024
  • Research Intern, VMware Research Certifier: a cryptographic certification protocol and framework for heterogeneous secure enclaves · with Dr. John Manferdelli Summer 2022
  • Founding Engineer, MobIQ With Prof. Yuanjie Li and Prof. Songwu Lu 2018 – 2020
  • Research Engineer, Siemens Gaussian-mixture time-series anomaly detection for industrial sensors Summer 2017
08 · Speaking

Invited Talks & Media

  • Sep 2025Towards Scalable Data Infrastructure for Humanoid Intelligence — Berkeley Humanoid Intelligence Center
  • Sep 2025Interviewee, "Mastering Research — Part 1/3: Choose Smart, Read Smarter!" — IEEE Robotics & Automation Magazine
  • Jun 2025Augmenting Open X-Embodiment Datasets — Google DeepMind–BAIR Commons
  • May 2025FogROS: A Cloud Robotics System — General Robotics
  • Feb 2025Cloud Robotics for Automation and Driving — Bosch Research
  • Jun 2024Cloud Robotics: Past, Present and Future — Berkeley Sky Seminar
  • May 2024FogROS2-SGC: Secure Global Connectivity — Berkeley Sky Retreat
  • Jan 2024Fog-RT-X: Cloud-Based Large-Scale Collection and Management — Google DeepMind
  • Aug 2023FogROS2: A Cloud Robotics Framework — Autodesk
  • Aug 2023Cloud Robotics and FogROS2 — ROS-by-the-Bay, Open Robotics
09 · Community

Teaching & Service

Teaching

  • Lead organizer, ICRA 2024 tutorial
    Cloud and Fog Robotics: A Hands-On Tutorial with ROS 2 and FogROS2
  • Graduate Student Instructor
    IEOR 115: Industrial and Commercial Data Systems (Fall 2024)
  • Mentor
    CS 262: Advanced Operating Systems (Fall 2020, Fall 2025)
  • Master's capstone advisor
    CMU MCDS, co-advised with Prof. Jeffrey Ichnowski (Fall 2023) · Berkeley MEng (Spring 2022)

Academic Service

  • Session Chair
    Multi-Robot Systems, ICRA 2025
  • Journal reviewer
    IEEE JSAC · IEEE Internet of Things Journal · IEEE RA-L
  • Conference reviewer
    ICRA 2023–2026 · IROS 2022, 2024, 2025 · CASE 2024, 2025 · ICCV 2025