Ziwei Liu

@liuziwei7

Associate Professor @NTUsg - Vision, Learning and Embodied AI | Building @ropedia_ai

Singapore

liuziwei7.github.io

Joined January 2018

1,698 Following

13.6K Followers

Replies Reposts Media

Ziwei Liu
@liuziwei7
Oct 25
🤖 Has GPT-6-Astra achieved embodied intelligence? We probe the limit of GPT-6-Astra across 12 benchmarks and 8 real-robot platforms 🥳 With the right harness and enough budget, GPT-6-Astra could *saturate most benchmarks* 😵 Real execution: not yet 🔗 sensenova.ai/has-gpt-6-astr…
Zhongang Cai
@caizhongang
Oct 19
03:03
12 Simulation benchmarks, 8 real robots, 1 question: Has GPT-6-Astra achieved Embodied Intelligence? Our answer: - Embodied competence? Yes. - Practical embodied inteligence? Not yet. Here’s what we found: sensenova.ai/has-gpt-6-astr… HF daily paper: huggingface.co/papers/2610.22…
3
Ziwei Liu
@liuziwei7
Oct 24
👋Making Physical State Explicit in Dexterous Manipulation World Models👋 #DeXplicit separates physical prediction from visual synthesis to generate physically grounded visual futures for dexterous manipulation. 📊 40% less point error ⚡️ 57× speedup 🔗 dexplicit.github.io
Adam Abecid
@AbecidAdam
Oct 23
1/7 Introducing DeXplicit: making physical state explicit in dexterous manipulation world models. Given an initial scene and a proposed hand trajectory, DeXplicit 3D forecasts how objects move, articulate and deform. The DeXplicit renderer turns that geometry into future video.
Ziwei Liu
@liuziwei7
Oct 22
🦾Generalist Dexterous Bimanual Tracking Control from Human Demonstrations🦾 We present #DONIC, a generalist bimanual tracker trained across diverse human-motion references 📊86.79% success on unseen trajectories 🎯57.32% success on held-out objects 🔗 ropedia.github.io/DONIC/
Ropedia
@ropedia_ai
Oct 22
1/7 Ropedia Super Week Day 3 Two hands. One policy. 🤲🤖 How do you teach robot hands not one skill, but every two-handed interaction humans show them? Today Ropedia releases DONIC: one policy that tracks human bimanual demos and generalizes to unseen trajectories and objects.
1
Ziwei Liu
@liuziwei7
Oct 21
🔥General coding agent beats frontier AI models on spatial reasoning 🚀Meet 𝐒-𝐀𝐠𝐞𝐧𝐭 𝟐.𝟎 🧠Think with code, not just intuition 👁️Turn visual scenes into measurable evidence 🏆Surpass GPT-6 Astra on MMSI & ReVSI-Bench 🏠ropedia.github.io/S-Agent-2.0/ 🧑‍💻github.com/Ropedia/S-Agen…
Ropedia
@ropedia_ai
Oct 21
1/6 Ropedia Super Week Day 3 Can a coding agent outperform GPT-6 on spatial reasoning—without additional training? Introducing S-Agent 2.0. With a stronger backbone, our framework surpasses GPT-6 on evaluated spatial reasoning tasks. Here's how we did it.
3
Ziwei Liu
@liuziwei7
Oct 20
😎Learning World-Grounded Egocentric Representations from Human Experience😎 #EgoXperience is a full-stack framework for jointly reconstructing camera motion, scene geometry, hands, and objects in a shared 3D world - Page: ropedia.github.io/EgoXperience/ - Code: github.com/Ropedia/EgoXpe…
Ropedia
@ropedia_ai
Oct 20
02:00
1/6 Ropedia Super Week Day 2 How can we turn first-person human experience into structured 3D spatial intelligence? For Day 2 of Ropedia Super Week, we're introducing two complementary research projects: HoliWorld and EgoXperience. Both explore how we can make first-person
2

Ziwei Liu

Ziwei Liu’s posts