Log inSign up
Yilin Wu
63 posts
user avatar
Yilin Wu
@YilinWu11
Robotics Ph.D. @CMU, intern at@Amazon FAR, prev MSCS @Stanford, prev intern at Nvidia
Pittsburgh, PA
yilin-wu98.github.io
Joined September 2020
327
Following
1,039
Followers

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • Pinned
    user avatar
    Yilin Wu
    @YilinWu11
    Jun 16
    So excited to share 𝚆𝙴𝙰𝚅𝙴𝚁 🌎, co-led with @arnavkj95 ! World models are becoming a powerful tool for robot learning — but for real robots, they need to be more than visually realistic. They also need to be consistent, efficient, and useful for decision-making. 🤖✨
    user avatar
    Arnav Jain
    @arnavkj95
    Jun 16
    🚀 Introducing 𝚆𝙴𝙰𝚅𝙴𝚁 🌎, a world model for robot manipulation built around three key desiderata: (i) 🔍 Fidelity, (ii) 🔗 Consistency, and (iii) ⚡ Efficiency. 𝚆𝙴𝙰𝚅𝙴𝚁 unlocks multiple downstream applications—policy evaluation, policy improvement, and test-time
    Image
    00:00
    12K
  • user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Ever watch your imitation-based robot policy do something bizarre? Wish you could fix it—no retraining needed? Meet FOREWARN, a VLM-in-the-loop system that steers multi-modal generative policies toward the right outcomes, on the fly.
    Image
    00:00
    65K
  • user avatar
    Yilin Wu
    @YilinWu11
    Jun 21, 2025
    I am at RSS this week. Tomorrow I will present our work FOREWARN at Reasoning for Robust manipulation workshop and FM4RoboPlan workshop. I will also give a talk at Imitation Learning Session I Sunday afternoon. Happy to catch up or chat about research during the conference!
    user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Ever watch your imitation-based robot policy do something bizarre? Wish you could fix it—no retraining needed? Meet FOREWARN, a VLM-in-the-loop system that steers multi-modal generative policies toward the right outcomes, on the fly.
    Image
    00:00
    8.9K
  • user avatar
    Yilin Wu
    @YilinWu11
    Jun 22, 2025
    I am going to give a talk about our work at 5:30pm-6:30pm today in Imitations Learning Session I. Also we have our poster at poster board 54. If you are interested in our work, feel free to check our poster and chat with me!
    Image
    Image
    01:10
    user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Ever watch your imitation-based robot policy do something bizarre? Wish you could fix it—no retraining needed? Meet FOREWARN, a VLM-in-the-loop system that steers multi-modal generative policies toward the right outcomes, on the fly.
    12K
  • user avatar
    Yilin Wu
    @YilinWu11
    Sep 24, 2024
    Happy to share our new paper ToolGen🛠️For generalizable multi-task tool-use skills, our method imagines the tool shape and uses flow to generate point cloud trajectory for manipulating dough with continuous contact. Our method is adaptable even with unseen real-world tools.👇
    user avatar
    Carl Qi
    @carl_qi98
    Sep 12, 2024
    How can an autonomous agent leverage novel tools to cut, roll, and scoop a piece of dough, given just a few tool shapes for training? Our method generates a “desired tool shape” that performs the motion and then matches the real tool to the generated shape.sites.google.com/view/toolgen
    Image
    GIF
    9K
  • user avatar
    Yilin Wu
    @YilinWu11
    Aug 2, 2025
    Congrats 🎊 Looking forward to seeing great work coming from your new lab
    user avatar
    Chenfeng_X
    @Chenfeng_X
    Aug 1, 2025
    📢 Excited to sharing a little late update (before it is no longer news): I’ll be joining @UTAustin @UTCompSci as an Assistant Professor! I'm recruiting PhD students from @UTCompSci in the Fall 2025 cycle and also looking for RAs/interns! More info see chenfengx.com
    Image
    15K
  • user avatar
    Yilin Wu
    @YilinWu11
    Jun 11, 2025
    Leveraging a world model and a learned reward function to optimize the test-time performance of the behavior cloning policies. Great to see more work in this direction—check out Gokul’s latest research!
    user avatar
    Gokul Swamy
    @g_k_swamy
    Jun 10, 2025
    Say ahoy to 𝚂𝙰𝙸𝙻𝙾𝚁⛵: a new paradigm of *learning to search* from demonstrations, enabling test-time reasoning about how to recover from mistakes w/o any additional human feedback! 𝚂𝙰𝙸𝙻𝙾𝚁 ⛵ out-performs Diffusion Policies trained via behavioral cloning on 5-10x data!
    Image
    GIF
    3.5K
  • user avatar
    Yilin Wu
    @YilinWu11
    Apr 27, 2025
    Check out our paper at the poster session at 8:00am! We also have an oral talk at 17:10 today! I cannot go but Michelle Zhao will present our work at the workshop. Feel free to come to our poster if you are interested in our work!
    2.7K
  • user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Replying to @YilinWu11
    1/ Quick links🔗: check out our paper (arxiv.org/abs/2502.01828) and project site (yilin-wu98.github.io/forewarn)!
    arXiv logo
    arxiv.org
    From Foresight to Forethought: VLM-In-the-Loop Policy Steering via...
    While generative robot policies have demonstrated significant potential in learning complex, multimodal behaviors from demonstrations, they still exhibit diverse failures at deployment-time....
    1.5K
  • user avatar
    Yilin Wu
    @YilinWu11
    Jul 18, 2024
    We will present HACMan++ at #RSS2024 from 8:30-9:30 am tomorrow (Friday) at the Manipulation session and at the poster session in room Senaatszaal. Come to check our poster if you are around!
    user avatar
    Bowen Jiang
    @bwww08
    Jul 14, 2024
    Introducing 𝐇𝐀𝐂𝐌𝐚𝐧++, accepted to RSS 2024! We’ve developed a 𝐜𝐨𝐧𝐭𝐚𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 action space for RL for manipulation. Our approach achieves: •Long-horizon contact reasoning •Generalization to diverse unseen objects •Effectiveness across diverse tasks (1/8)
    Image
    00:00
    490
  • user avatar
    Yilin Wu
    @YilinWu11
    Jul 14, 2024
    Want a new solution for long-horizon contact-rich manipulation tasks?Our paper shows with a generic set of spatially-grounded primitives, we can achieve that with long-horizon contact reasoning for a wide variety of objects and tasks! Check out our paper at RSS next week.
    user avatar
    Bowen Jiang
    @bwww08
    Jul 14, 2024
    Introducing 𝐇𝐀𝐂𝐌𝐚𝐧++, accepted to RSS 2024! We’ve developed a 𝐜𝐨𝐧𝐭𝐚𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 action space for RL for manipulation. Our approach achieves: •Long-horizon contact reasoning •Generalization to diverse unseen objects •Effectiveness across diverse tasks (1/8)
    Image
    00:00
    386
  • user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Replying to @YilinWu11
    8/ Want to learn more? Check our paper (arxiv.org/abs/2502.01828) and project site (yilin-wu98.github.io/forewarn). Big thanks to my amazing collaborators @tianran_, @g_k_swamy, and my advisor @andrea_bajcsy for their invaluable work and continuous support!
    arXiv logo
    arxiv.org
    From Foresight to Forethought: VLM-In-the-Loop Policy Steering via...
    While generative robot policies have demonstrated significant potential in learning complex, multimodal behaviors from demonstrations, they still exhibit diverse failures at deployment-time....
    863
  • user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Replying to @YilinWu11
    3/ Key insight: treat runtime policy steering as a stochastic MPC problem over low-level action plans. In open-world settings, physics-based models and explicit rewards are hard to define so we use a learned world model for prediction and a VLM for open-ended evaluation instead.
    Image
    00:00
    943
  • user avatar
    Yilin Wu
    @YilinWu11
    Feb 17, 2025
    Replying to @YilinWu11
    2/ Although state-of-the-art imitation policies learn from demonstrations, they often degrade during deployment, from total failures to subtle misalignments. How can we correct them—without retraining, given that it is capable of generating correct behaviors sometimes?
    Image
    00:00
    1.1K
This post is unavailable.
Advertisement
Advertisement