Twitxy Saved

Mike Shou

Verified account This account is verified. Learn more

@MikeShou1

Assoc Prof at NUS. Forbes 30 under 30 Asia. Previously at Facebook AI and Columbia U. Passionate about video, multi-modal, AI assistant.
sites.google.com/view/showlab Joined December 2020
777 Following 3K Followers 331 Posts

Latest posts 5

Pinned post

Mike Shou Verified account @MikeShou1
(1/6) X-Humanoid πŸ€–: Scaling up data for Humanoid Robots.

We convert human daily activity videos (from Ego-Exo4D) into humanoid videos (i.e., Tesla Optimus) performing tasks like cooking or fixing a bike.

This data can be potentially used to train robot policies and world models. πŸ”₯

Project page: showlab.github.io/X-Humanoid/
Paper link: arxiv.org/abs/2512.04537
0:12
24 replies 68 reposts 417 likes 89K views View on X
Mike Shou Verified account @MikeShou1
Heartiest congrats to Jinheng, very well deserved! Thanks to Google for the recognition!
Jinheng Xie @Sierkinhane1 4d
Thrilled to receive the 2026 Google PhD Fellowship! Huge thanks to my advisor @MikeShou1 and my collaborators for their support throughout my PhD, and to Amir, @JunhaoZHANG19, @neal_wadhwa, and @Jimantha for a wonderful internship at Google in NYC. Thank you, Google!
View quoted post on X
0 replies 1 reposts 17 likes 1.7K views View on X
Mike Shou Verified account @MikeShou1
A survey on Multimodal Embodied Agents, centering around "Perceive β†’ Anticipate β†’ Plan β†’ Act β†’ Verify" -- a unified capability-centric framework for understanding multimodal agents and robotic systems. πŸ‘‡
Anno Yanzhe Chen Verified account @Anno_YanzheChen Sep 11
[2/5] We introduce PAPAV:

Perceive β†’ Anticipate β†’ Plan β†’ Act β†’ Verify

A unified capability-centric framework for understanding multimodal agents, robotic systems, and their convergence into Multimodal Embodied Agents.
View quoted post on X
1 replies 0 reposts 10 likes 1.4K views View on X
Mike Shou Verified account @MikeShou1
🦾 GPT Astra for robot is impressive.. but slow! Can we use light-weight Qwen for real-time control, or use any general VLMs like Gemini, Claude?

πŸ•ΉοΈ We are thrilled to introduce Show-Harness, a semantic action interface that brings the physical intelligence from a general VLM agent into robot-use. No more calibration. No extra VLA or SAM.

🧠 Show-Harness generalizes across tasks and scenes, using GPT, Gemini, Claude zero-shot without training, or just fine-tuning Qwen with 164 demos within 1 hour.

Model and codeπŸ‘‡
🌐 Project Page: showlab.github.io/Show-Harness/
πŸ“„ Paper: arxiv.org/abs/2609.10522
πŸ’» Code: github.com/showlab/Show-H…
πŸ€— Model & Data: huggingface.co/collections/sh…
Zechen Bai Verified account @ZechenBai Sep 10
[1/8] 🦾 Could a strong robot policy already be hiding inside your VLM, not just GPT Astra, but also Gemini, Claude, even light-weight Qwen and InternVL?
πŸ•ΉοΈ We are thrilled to introduce Show-Harness, an Embodied Harness that lets VLMs play real robotsβ€”like Claude Code for robotics.
No more calibration. No extra VLA.
View quoted post on X
3 replies 17 reposts 116 likes 11K views View on X
Mike Shou Verified account @MikeShou1
After pivoting to embodied AI last year, this is our first time submitting to #CoRL 2026, and I’m excited to share that Show Lab has 3 out of 3 papers accepted! πŸŽ‰

Huge congratulations to the students behind these papers β€” all were 1st-year PhD or master students, completely new to the field, who spent tremendous efforts to pick up the know-how of robot learning πŸ€–

1. Supervise What Survives: Geometry-Guided VLA Adaptation from Synthetic Robot Videos

How robot can learn from human demo video? Video gen model can edit human video as robot video, but how to use these generated human-edit-as-robot video for training robot?

Instead of hoping to learn how to control, we propose to guide robot learning with the geometry in the human-edit-as-robot video.

2. Where Success Breaks: Failure-Boundary Learning for Robust Vision-Language-Action Models

When training VLAs, we often focus on successful executions. But failures contain equally important information about where a robot is starting to go wrong.

We introduce Failure-Boundary Learning, reframing robust VLA adaptation as the problem of discovering, localizing, and shaping the boundary between recoverable deviations and task failure.

3. From Video Frames to Robot Meta-Frames: A Unified Latent Interface for World Action Models

World Action Models such as Cosmos Policy show great potential for jointly predicting actions and video. However, different robot signals β€” images, proprioception, actions, future states, and values β€” are typically represented as separate images, making inference computationally costly.

We introduce MetaWAM, which packs these different modalities into a single unified meta-frame in the latent space. On RoboCasa, MetaWAM achieves 68.9% success, while reducing inference latency by 39%.