Yu Haibao

Towards Physical AGI.
PhD@HKU-MMLAB.

[Yu Haibao @walk2out]

@qx_dong impressive work.

[Yu Haibao @walk2out]

Introducing PhysLadder!

A hierarchical benchmark for GPT-6 Astra and other general-purpose models in contact-rich manipulation.

16 tasks. 3 levels. Physically and visually matched simulation and real-world environments.

https://t.co/t2aU12sTL5

[Yu Haibao @walk2out]

@JiankaiSun

[Yu Haibao @walk2out]

Introducing PhysLadder!

A hierarchical benchmark for GPT-6 Astra and other general-purpose models in contact-rich manipulation.

16 tasks. 3 levels. Physically and visually matched simulation and real-world environments.

https://t.co/t2aU12sTL5

[Yu Haibao @walk2out]

GPT-6 Astra, zero-shot: 96 trials across 16 tasks and two environments. Success from L1 to L3: Real: 80.0% → 47.6% → 0% Sim: 66.7% → 47.6% → 0% Local progress did not reliably add up to task completion. No L3 task was completed in this evaluation.

[Yu Haibao @walk2out]

Sim-real agreement: 10/16 tasks had identical success counts; the other 6 differed by one success.

Shared difficulties support matched simulation as a repeatable testbed for Astra.

Paper: https://t.co/0UxJmgigxq
Code: https://t.co/DXEltRVg7h

[Yu Haibao @walk2out]

Three levels of physical interaction:

L1: Contact selection and geometric alignment.
L2: Force regulation during sustained contact.
L3: Coordinated contact reconfiguration and force regulation.
From placing a pen in a cup to unscrewing a cap and sweeping blocks.

[Yu Haibao @walk2out]

Thanks for sharing our GaussianDream~

[Robots Digest 🤖 @robotsdigest]

GaussianDream argues current Vision-Language-Action (VLA) robot models mainly imitate actions from videos, but do not explicitly model how the environment will change after interaction. Existing 3D VLAs add geometry like depth or point clouds, yet mostly capture only the current scene. World models can predict futures, but usually rely on expensive video rollouts or latent simulations that are too slow for real-time robotic control.

[Yu Haibao @walk2out]

@xie_yaqi @UofIllinois Congratulations, Yaqi!

[Yu Haibao @walk2out]

🌐Workshop: https://t.co/7XUv1fkC8w
🏆 Challenge: https://t.co/etcutuwLa6

[Yu Haibao @walk2out]

#IROS2026 Call for participants 📢
Excited to announce our IROS 2026 Workshop: Physical World Models for Scaling Embodied AI, together with the WorldArena 2.0 Challenge ($7,000 prize pool)!
The workshop is held on October 1 in Pittsburgh.

[Yu Haibao @walk2out]

🤝 Co-organized with @Haibao Yu @Dandan Zhang @Lei Yang @Weitao Zhou @Jianing Qiu @Jiankai Sun @Tianxing Chen @Lan Wei @Wen Fan @Walter Zimmer @Chen Gao @Yong Li @Ping Luo

[Yu Haibao @walk2out]

🎤 Keynote speakers: @Jiajun Wu @Hongyang Li @Rudra P.K. Poudel @Abhinav Valada @Ding Zhao @Katerina Fragkiadaki

[Yu Haibao @walk2out]

Welcome to submit your paper on our workshop!

[Jiankai Sun @JiankaiSun]

🚀 Call for Papers | Safe World Models for Trustworthy Embodied AI #ECCV 2026

World models are transforming embodied AI—but can we trust them in the real world?

📍 ECCV 2026, Malmö
📅 Deadline: July 15, 2026
🔗 https://t.co/Z1QROFfCPI

#ECCV2026 #EmbodiedAI #WorldModels #SafeAI

[Yu Haibao @walk2out]

@robotsdigest Thanks for sharing my team’s work

[Yu Haibao @walk2out]

@Figure_robot Great Demo

[Yu Haibao @walk2out]

@peteflorence Curious about the architectural leap in GEN-1, as it seems to bypass the conventional VLA and World Model trajectories.

[Yu Haibao @walk2out]

Impressive breakthroughs

[Generalist @GeneralistAI]

Introducing GEN-1.

Our latest milestone in scaling robot learning.

We believe it to be the first general-purpose AI model to master simple physical tasks.

99% success rates, 3x faster speeds, adapts in real time to unexpected scenarios, w/ only 1 hour of robot data.

More🧵👇

[Yu Haibao @walk2out]

Excellent Work

[OpenDriveLab @OpenDriveLab]

🚀 Introducing SMASH — the world’s first high-dynamic humanoid robot for outdoor table tennis, fully autonomous with onboard perception.

No motion capture needed—SMASH uses only onboard sensors for stable, full-body human–robot interaction in real-world settings.

Embodied intelligence is stepping out of the lab.

🔗https://t.co/xphEg4HuoG

#HumanoidRobots #TableTennis #RobotLearning #SportsTech #AI #OpenDriveLab #Algorithms

[OpenDriveLab @OpenDriveLab]

🚀 Introducing SMASH — the world’s first high-dynamic humanoid robot for outdoor table tennis, fully autonomous with onboard perception.

No motion capture needed—SMASH uses only onboard sensors for stable, full-body human–robot interaction in real-world settings.

Embodied intelligence is stepping out of the lab.

🔗https://t.co/xphEg4HuoG

#HumanoidRobots #TableTennis #RobotLearning #SportsTech #AI #OpenDriveLab #Algorithms