Guangqi Jiang
@LuccaChiang
Followers
510
Following
2K
Media
17
Statuses
163
MSCS @UCSanDiego | Prev. RA @Tsinghua_IIIS | Robot Learning, Embodied AI, and Vision
San Diego, CA
Joined September 2023
Ever want to enjoy all the privileged information in sim while seamlessly transferring to the real world? How can we correct policy mistakes after deployment? ๐Introducing GSWorld, a real2sim2real photo-realistic simulator with interaction physics with fully open-sourced code.
6
65
274
Really appreciate the recognition and the chance to be part of this project. Iโm currently seeking the PhD positions for the coming year โ Iโd be glad to connect with labs or researchers in robotics and related fields! Personal website ๐
And also appreciate @laiiweiโs contribution to this work. He is highly technical in real-robot experiments and works independently. He will be on the PhD market in the coming year ๐ฉ
0
1
6
Glad we have PI by our side.
We just released results for our newest VLA from Physical Intelligence: ฯ*0.6. This one is trained with RL, and it makes it quite a bit better: often doubles throughput, enables real-world tasks like folding real laundry and making espresso drinks at the office.
0
0
18
Thank you for featuring our work!
GSWorld: Closed-Loop Photo-Realistic Simulation Suite for Robotic Manipulation Project:ย https://t.co/lz2zgphUUp Paper:ย https://t.co/tImAwZvxrc This new work explores the idea to combine 3D Gaussian Splatting with physics engine to provide a more photo-realistic simulation for
1
0
11
We've been seeing precise manipulation these days, but most of them are limited to grippers or low-dof hands. Here, we propose a method to make high-dof dexterous hands more powerful in manipulation. ๐ ๐บ-๐๐ฐ๐ฎ๐น๐ฒ ๐๐๐ก๐-๐ด๐ฟ๐ฎ๐ถ๐ป๐ฒ๐ฑ ๐ฝ๐ฟ๐ฒ๐ฐ๐ถ๐๐ถ๐ผ๐ป!๐
How do we make dexterous hands handle both power and precision tasks with ease? ๐ซณ๐๐ซฐ We introduce Power to Precision (๐ชโก๏ธ๐ฏ), our new paper that optimizes both control and fingertip geometry to unlock robust manipulation from power grasp to fine-grained manipulations. With
2
2
26
How do we make dexterous hands handle both power and precision tasks with ease? ๐ซณ๐๐ซฐ We introduce Power to Precision (๐ชโก๏ธ๐ฏ), our new paper that optimizes both control and fingertip geometry to unlock robust manipulation from power grasp to fine-grained manipulations. With
5
53
173
Real2sim should always be the prior to sim2real. The results look really promising! ๐Also checkout our similar paper with a real2sim2real pipeline here!
We release OpenReal2Sim, an open-source toolbox for real-to-sim reconstruction and robot simulation. A key difference from prior work is our focus on building an interactive digital twin from in-the-wild data โ even Internet images or generated videos. Try it out: Interactive
1
7
56
Due to a large embodiment gap, physics awareness is crucial for manipulation tasks, especially contact-rich ones.
๐ธ๏ธ Introducing SPIDER โ Scalable Physics-Informed Dexterous Retargeting! A dynamically feasible, cross-embodiment retargeting framework for BOTH humanoids ๐ค and dexterous hands โ. From human motion โ sim โ real robots, at scale. ๐ Website: https://t.co/ieZfG2Q4L0 ๐งต 1/n
0
1
9
Wow itโs data-free learning.
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
0
0
2
Real2sim evaluation is very useful, cheap, scalable, remote-friendly, etcโ
. Robot learning researchers will get much more productive if we can know the policy performance w/o being on site. Also, check out our recently released GSWorld with similar ideas https://t.co/Ri34MMndtq.
๐งต Evaluating robot policies in the real world is slow, expensive, and hard to scale. During my internship at @SceniXai this summer, we had many discussions around the two key questions: how accurate must a simulator be for evaluation to be meaningful, and how do we get there?
0
5
23
๐งต Evaluating robot policies in the real world is slow, expensive, and hard to scale. During my internship at @SceniXai this summer, we had many discussions around the two key questions: how accurate must a simulator be for evaluation to be meaningful, and how do we get there?
8
28
139
Excited to introduce TWIST2, our next-generation humanoid data collection system. TWIST2 is portable (use anywhere, no MoCap), scalable (100+ demos in 15 mins), and holistic (unlock major whole-body human skills). Fully open-sourced: https://t.co/fAlyD77DEt
20
108
453
Scaling law is a universal belief.
Introducing GEN-0, our latest 10B+ foundation model for robots โฑ๏ธ built on Harmonic Reasoning, new architecture that can think & act seamlessly ๐ strong scaling laws: more pretraining & model size = better ๐ unprecedented corpus of 270,000+ hrs of dexterous data Read more ๐
0
0
3
We have seen so many VLMs nowadays, but do they show vision-language cross-modality ability? You might want to take a look at the comprehensive evaluations from the benchmarking here. Also informative for robotic VLA models.
Unified multimodal models can generate text and images, but can they truly reason across modalities? ๐จ Introducing ROVER, the first benchmark that evaluates reciprocal cross-modal reasoning in unified models, the next frontier of omnimodal intelligence. ๐ Project:
1
0
22
Unified multimodal models can generate text and images, but can they truly reason across modalities? ๐จ Introducing ROVER, the first benchmark that evaluates reciprocal cross-modal reasoning in unified models, the next frontier of omnimodal intelligence. ๐ Project:
5
30
192
๐จ Paper Alert ๐จ โก๏ธPaper Title: GSWorld: Closed-Loop Photo-Realistic Simulation Suite for Robotic Manipulation ๐Few pointers from the paper ๐ฏAuthor of this paper presented โGSWorldโ, a robust, photo-realistic simulator for robotics manipulation that combines 3D Gaussian
1
2
4
GSWorld is a photo-realistic simulator combining 3D Gaussian Splatting with physics engines to reduce sim2real visual gap, with diverse downstream applications.Creating a digital twin is getting closer to reality, but how can we maximize its utility for robotic manipulation?
0
1
2
GSWorld: Closed-Loop Photo-Realistic Simulation Suite for Robotic Manipulation Contributions: โข A solid real-to-sim-to-real pipeline: Our robust pipeline accurately aligns the simulation with the real environment, enabling a wide range of subsequent applications. โข Simulation
1
10
58
Thank you for your early post, Mr. NeRF. You are even faster than our release post here https://t.co/SulJAV1RES. Please kindly refer to our website for the updated YouTube video and GitHub codebase.๐
GSWorld: Closed-Loop Photo-Realistic Simulation Suite for Robotic Manipulation Contributions: โข A solid real-to-sim-to-real pipeline: Our robust pipeline accurately aligns the simulation with the real environment, enabling a wide range of subsequent applications. โข Simulation
0
0
5
I am writing related work section for a paper and would like to learn what is going on in the field ๐ซฐ"dexterous hand" and ๐ค "dexterous manipulation", get 1k papers from arXiv in 5 mins and use 5 mins generate the following survey : )
As an AI researcher, are you interested in tracking trends from CV/NLP/ML to roboticsโeven Nature/Science. Our paper โReal Deep Research for AI, Robotics & Beyondโ automates survey generation and trend/topic discovery across fields ๐ฅExplore RDR at https://t.co/uBJ0bYNy3R
1
1
3
Setting up diverse scenes in simulation for robotics is often tedious โ but real2sim makes it seamless. Excited to see how this line of work scales sim2real training and brings rigorous benchmarks for evaluating progress in robotic manipulation.
Ever want to enjoy all the privileged information in sim while seamlessly transferring to the real world? How can we correct policy mistakes after deployment? ๐Introducing GSWorld, a real2sim2real photo-realistic simulator with interaction physics with fully open-sourced code.
1
3
27