Yishu Li

Incoming CS PhD @berkeley_ai, MSR @CMU_Robotics, Prev CS Undergrad @Tsinghua_Uni

[Yishu Li @LisaYishu]

A closed door looks the same whether it pushes or pulls. Two identical-looking boxes might have different center of mass. How should robots act when a single visual observation isn't enough?

Introducing HAVE 🤖, our method that reasons about past interactions online! #CORL2025

[Yuanchen Ju @ju_yuanchen]

#ICML2026Spotlight Standard diffusion is powerful, but inference latency is a real headache for real-time robot control. HDFlow uses rectified flow for instant trajectory generation in long-horizon tasks. Fantastic work by @girish_432 @AveryJuuu0213 !

[Sriram Krishna ✈️ RSS 2026 @sriramsk99]

Boo! 👻 GHOST has been accepted to #RSS2026!

Can we learn manipulation skills from human video (without action retargeting)?

Yes! GHOST learns generalizable manipulation skills by training a hierarchical policy with an embodiment-agnostic sub-goal predictor + embodiment-specific controller.

🧵(1/7)

[Ellina Zhang @EllinaZhang]

🧩 #ICML2026 💥 How can a model discover the 3D objects in a scene—their shape, color, and position—without any labels? Introducing 3D-DLP, a self-supervised object-centric model that decomposes colored 3D scenes (RGB-D and voxels) into a set of 3D latent particles.

[Yixuan Huang @YixuanHuang13]

Meet KinDER — a stress test for robot physical reasoning. All 13 methods failed 😈

🌎 25 environments
♾️ Infinite tasks
🏋️ Gymnasium API
⚒️ Over 20 parameterized skills
🪧 Human demonstrations
📊 13 baselines (planning and learning)

From @Princeton @CMU_Robotics @ICatGT @CambridgeMLG @nvidia @MIT_CSAIL

🧵 1/n

[Yishu Li @LisaYishu]

@hq_fang @Stanford @StanfordSVL @drfeifei congrats Haoquan!

[Yishu Li @LisaYishu]

Great work, Yuanchen!

[Yuanchen Ju @ju_yuanchen]

🤖For embodied agents in household environments, we tackle two fundamental questions: 1️⃣ What is the optimal scene representation? 2️⃣ Can a VLM leveraging this representation actually improve spatial understanding and task planning?

Introducing MomaGraph: State-Aware Unified Scene Graphs with Vision-Language Models for Embodied Task Planning. 👉: https://t.co/KK4YCknG1K and 🔗:https://t.co/TfeZwIHMkE

Key Ideas：
MomaGraph jointly models spatial AND functional relationships with part-level interactive nodes.
MomaGraph is designed to be: ✅ Task-Relevant: Filters visual noise to keep only what matters for the instruction. ✅ Dynamic & State-Aware: MomaGraph adapts. 🔄 It explicitly models object states and dynamic changes in the environment.

We built MomaGraph to bridge the gap between the Spatial VLM and Robotics communities. 🌉 Our hope is that this work serves as a foundation for the next generation of intelligent, adaptive embodied agents. 🦾✨Questions and feedback welcome. 🚀
#Robotics #EmbodiedAI #CV #LLM #SceneGraph

[Li Yi @ericyi0124]

Thrilled to share one of my favorite works this year: DexNDM!

We bridge the Sim2Real gap for dexterous in-hand rotation, achieving a true "0-to-1" advancement. The key? DexNDM learns from biased, real-world data without needing any successful demonstrations.

Now a general-purpose dexterous hand can stably rotate large books, long rods, & complex objects around any axis, from any wrist pose. This powerful primitive enables complex, long-horizon tasks like teleoperated screwing and furniture assembly.

📄 Paper: https://t.co/KR3m464V6l
🌐 Project Page: https://t.co/rHtV50RvBN

[Alexis Hao @hao_alexis]

Introducing FMVP: a method that adapts to natural arm motions during robot-assisted dressing.

Pre-trained on vision in sim, fine-tuned with limited real-world vision+force data, and tested in a 12-user, 264-trial study, FMVP is robust across garments and motions. #CoRL2025

[Divyam Goel @divyamgo10]

How do we discover a robot's failure modes before deploying it in the real world? Standard benchmarks often don't capture the full picture, leaving policies vulnerable to plausible variations in object shape.

Thrilled that our work, "Geometric Red-Teaming for Robotic Manipulation," has been accepted as an oral presentation at #CoRL2025! We introduce a framework to automatically find these geometric blindspots.
https://t.co/mh84bi7Wbp
🧵

[Yishu Li @LisaYishu]

Check out @_ksaha's very cool work on learning and planning for long-horizon tasks!

[Kallol Saha @_ksaha]

🚨Introducing SPOT: Search over Point Cloud Object Transformations. SPOT is a combined learning-and-planning approach that searches in the space of object transformations.

Website: https://t.co/VwiyDI15FC
Paper: https://t.co/GmdF7hN0FG
Code: https://t.co/YC5sO9OZ0K

[Yishu Li @LisaYishu]

@_ksaha Thanks!!!

[Yishu Li @LisaYishu]

@AveryJuuu0213 Thanks congrats to you toooooooo

[Yishu Li @LisaYishu]

Come check out our poster at Poster Session 1 on Sept 28 @ 4:30 - 6:00PM at #CoRL2025!

Website: https://t.co/znEWXekoFo
Paper: https://t.co/oCMic1lHx6
Code: https://t.co/qlqpWdzaj3

Thanks to collaborators: @shhmxy2, @Ying_yyyyyyyy, @ktsim01, @BenAEisner, and @davheld.

[Yishu Li @LisaYishu]

A closed door looks the same whether it pushes or pulls. Two identical-looking boxes might have different center of mass. How should robots act when a single visual observation isn't enough?

Introducing HAVE 🤖, our method that reasons about past interactions online! #CORL2025

[Yishu Li @LisaYishu]

We analyzed how the number of action proposals provided to the verifier affects performance. The method is sample efficient, as performance improves significantly when the verifier is given just 5 samples. This demonstrates the efficiency of using a verifier for action selection.