sombit_d

70 posts

sombit_d

sombit_d

@DSombit

Robotics @INSAITinstitute, @ETH_en,@IITKgp Personal Website: https://t.co/zoZcKoySLs

Indian Institute of Technology Entrou em Mart 2018
1.4K Seguindo66 Seguidores
Oier Mees
Oier Mees@oier_mees·
Excited to share that I've started teaching a new course on "Robot Learning: From Fundamentals to Foundation Models" as an External Lecturer at ETH Zurich! I hope this course inspires students to see that Robot Learning is one of the most exciting fields to be working on today!
Oier Mees tweet media
English
17
19
439
16.7K
Tony Zhao
Tony Zhao@tonyzzhao·
Today, we present a step-change in robotic AI @sundayrobotics. Introducing ACT-1: A frontier robot foundation model trained on zero robot data. - Ultra long-horizon tasks - Zero-shot generalization - Advanced dexterity 🧵->
English
434
650
5.4K
2M
INSAIT Institute
INSAIT Institute@INSAITinstitute·
🔥 We’re releasing SPEAR-1 (spear.insait.ai) - a new robotic AI foundation model that achieves state-of-the-art performance with 20× less robotic data 🧠 Why it matters:SPEAR-1 is like the ChatGPT for robots - a single model that can perform many tasks, on any robot, in any environment. 💡 What’s new: unlike others, SPEAR-1 learns from both robotic and non-robotic 3D data, breaking the data bottleneck that slows robotic AI. 🤖 Open-weight, general-purpose, and multilingual for robots - a major step toward scalable robot learning. #Robotics #FoundationModels #3DPerception #Manipulation #INSAIT #Europe #DataEfficiency
INSAIT Institute tweet media
English
2
5
27
17.5K
sombit_d
sombit_d@DSombit·
270K hours of robot data ~30yrs. 90 data collectors working for 8hrs a day for year? Can be done in around 2M USD with UMI based ? P.S. Thats around 30B images at 30fps , ~15PB of image storage. Makes sense to lay new internet cables to transfer data (awesome stuff!!!)
Generalist@GeneralistAI

Introducing GEN-0, our latest 10B+ foundation model for robots ⏱️ built on Harmonic Reasoning, new architecture that can think & act seamlessly 📈 strong scaling laws: more pretraining & model size = better 🌍 unprecedented corpus of 270,000+ hrs of dexterous data Read more 👇

English
0
0
4
113
Guangqi Jiang
Guangqi Jiang@LuccaChiang·
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? 👉Introducing GSWorld, a real2sim2real photo-realistic simulator with interaction physics with fully open-sourced code.
English
7
67
294
79.5K
Murtaza Dalal
Murtaza Dalal@mihdalal·
Proudest moment of my research career, Neural MP won Best Student Paper at IROS! We built a fast, reactive neural motion planner that works in the real world and ended up being used to be enable incredibly long-horizon manipulation tasks in ManipGen (mihdalal.github.io/manipgen/). This was a culmination over a year's worth of effort starting all the way in Spring 2023, with a super fun discussion with @pathak2206 :) Neural MP would not have been possible with out the incredible work put in by the entire team, especially @Jiahui_Yang6709 (who is applying for PhD Fall 2026 and would be a great candidate for any program!) Definitely check out our robot videos, code and paper here: mihdalal.github.io/neuralmotionpl…
Deepak Pathak@pathak2206

Incredible news. Neural MP has won the Best Student Paper award at IROS 2025!! Congratulations to @mihdalal & @Jiahui_Yang6709 for leading the project along with @mendonca_rl, youssef, @rsalakhu. Neural MP is a major step in making motion planning end-to-end, fast & reactive.

English
18
14
149
15.1K
sombit_d
sombit_d@DSombit·
@YouJiacheng They also study and ablate on different gradient magnitudes and schedules between noise and x, apart from removing time-conditioning
English
0
0
2
930
You Jiacheng
You Jiacheng@YouJiacheng·
Wait a minute, it seems that EqM ≈ Flow Matching without time conditioning? And kaiming's work arxiv.org/abs/2502.13129 already found that FM w/o time conditioning has a better FID. A rebranding?
Yilun Du@du_yilun

Excited to share Equilibrium Matching (EqM)! EqM simplifies and outperforms flow matching, enabling strong generative performance of FID 1.96 on ImageNet 256x256. EqM learns a single static EBM landscape for generation, enabling a simple gradient-based generation procedure.

English
5
30
342
45.3K
sombit_d retweetou
Yiting Chen
Yiting Chen@YitingChen07·
“You can’t make progress until you are able to measure it. Robotics still doesn’t have such a rallying call. No one agrees on anything.” I 💯 agree with the recent post from @DrJimFan. To break this impasse, we are excited to announce ManipulationNet (manipulation-net.org), a community-driven global infrastructure supported by @NIST, that enables benchmarking real-world robot manipulation research at scale with any robot at any time and anywhere on standardized task setups. ManipulationNet’s key features include: ✅delivers standardized hardware kits globally to support reproducible task setups ✅provides online task instructions in real-time, which could involve visual/language prompts, task-specific instructions, etc.. ✅collects authentic manipulation performance for comparable results on global leaderboards 🌍Join the ManipulationNet, let us aim to "connect the dots" of the up-to-date progresses and challenges to construct a network of real-world robot abilities and skills, a network of research roadmaps, and a network of research questions. Want to learn more about ManipulationNet? Check out the links below: 🔗Project website: manipulation-net.org 🔗Founding Committee and Developer Team: manipulation-net.org/committee.html 🔗Paper: manipulation-net.org/MNet_preprint.… (1/7)
Yiting Chen tweet media
English
9
22
109
26.7K
sombit_d retweetou
The Nobel Prize
The Nobel Prize@NobelPrize·
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2025 #NobelPrize in Physics to John Clarke, Michel H. Devoret and John M. Martinis “for the discovery of macroscopic quantum mechanical tunnelling and energy quantisation in an electric circuit.”
The Nobel Prize tweet media
English
403
7.5K
17.9K
4.7M
C Zhang
C Zhang@ChongZitaZhang·
After half a year I found the only thing I can change on my CV is my email address🤡
English
2
0
42
4.1K
sombit_d
sombit_d@DSombit·
@chris_j_paxton I just assumed a cheeky dig at authors moving to Meta 😅🤣
English
0
0
1
32
sombit_d
sombit_d@DSombit·
Lot of potential use cases!
NVIDIA AI@NVIDIAAI

With 3M+ downloads and counting, NVIDIA Cosmos is redefining physical AI. Announced at #CORL25, new Cosmos updates are allowing developers to generate diverse data for accelerating training robot models at scale. 👏 Cosmos Predict 2.5 will combine three models into one powerful model—reducing complexity, powering up to 30s video generation, and enabling multi-view simulations. 👏 Cosmos Transfer 2.5 will be 3.5x smaller yet faster and sharper—generating photorealistic synthetic data from 3D scenes or spatial inputs. 🔗 nvda.ws/48lW6eT

English
0
0
0
13
sombit_d retweetou
Séb Krier
Séb Krier@sebkrier·
According to this paper, the Unitree G1 humanoid robot secretly and continuously sends sensor and system data to servers in China without the owner's knowledge or consent. arxiv.org/abs/2509.14139
Séb Krier tweet media
English
118
284
1.7K
802.1K
clem 🤗
clem 🤗@ClementDelangue·
Something that's missing in robotics AI is the reflex from researchers and builders to share not only a video demo but also the code, datasets, policies, models or research papers for others to benefit from it and to show that what they did is not fake, staged or cherry-picked and can be replicated. That's something that we will work on @huggingface as I think it can really accelerate the field thanks to open-source collaboration & empower millions to become robotics AI builders. How can we help on this?
English
31
21
213
25.9K
Takahiro Miki
Takahiro Miki@ki_ki_ki1·
Update: I joined @GoogleDeepMind as research scientist last week! Excited to continue working on robotics there!
English
20
10
824
61.4K
sombit_d
sombit_d@DSombit·
@mzubairirshad Awesome work! Loved the emphasis on statistically significant evals
English
0
0
1
307
Zubair Irshad
Zubair Irshad@mzubairirshad·
🚀Thrilled to share what we’ve been building at TRI over the past several months: our first Large Behavior Models (LBMs) are here! I’m proud to have been a core contributor to the multi-task policy learning and post-training efforts. At TRI, we’ve been researching how LBMs can help robots learn faster, better, and more efficiently. The key takeaways: ✅ We built an evaluation pipeline to benchmark LBM performance with real 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 ✅ Pre-training on hundreds of tasks makes models more robust—plus, we can teach new, complex tasks with 80% 𝐥𝐞𝐬𝐬 𝐝𝐚𝐭𝐚 ✅ The bigger and more diverse the pre-training, the better the results Check out our overview video, webpage and paper for more details: ✨youtube.com/watch?v=DeLpnT… 🌎 toyotaresearchinstitute.github.io/lbm1/ 📄 arxiv.org/pdf/2507.05331 We hope this work helps move the field of robotics forward!
YouTube video
YouTube
Russ Tedrake@RussTedrake

TRI's latest Large Behavior Model (LBM) paper landed on arxiv last night! Check out our project website: toyotaresearchinstitute.github.io/lbm1/ One of our main goals for this paper was to put out a very careful and thorough study on the topic to help people understand the state of the technology, and to share a lot of details for how we're achieving it. youtube.com/watch?v=BEXFnr…

English
3
27
185
20.1K
Kyle🤖🚀🦭
Kyle🤖🚀🦭@KyleMorgenstein·
@chris_j_paxton I’ve been so confused by this absurd push across the industry to get to mass production. Virtually none of the hardware is mature enough for that yet and nothing sucks worse than having to support bad hardware. Imagine if Boeing acted this way with airplanes…
English
6
0
16
826
Chris Paxton
Chris Paxton@chris_j_paxton·
Generalist robots are hard, and there are consequences to just going for the human form factor without thinking about why and what you actually want it to do. But I am confident tesla can improve their design
The Humanoid Hub@TheHumanoidHub

Tesla has paused Optimus production to revise its design – the redesign could take around two months, according, per Chinese outlet LatePost Auto. Musk's mass production goal for this year is basically out of reach. According to sources in China’s supply chain, Tesla is making focused adjustments to the hardware and software design of its humanoid robot Optimus, and paused component procurement for the robot about two weeks ago. Two suppliers confirmed that Tesla has not explicitly canceled component orders but will only confirm a new mass production schedule and resume procurement after the design revision is complete. The redesign could take around two months. As of the end of May, Tesla had purchased enough parts to build 1,200 Optimus units and had produced nearly 1,000. Musk previously promised to produce 5,000 units this year. With the halt in component procurement, both sources believe that goal is now unlikely to be achieved. This adjustment started in early June with the departure of Milan Kovac, the original head of the Optimus project. According to one supplier, Tesla’s VP of AI Software Ashok Elluswamy has taken over the Optimus project and wants to improve the design before ramping up production. According to Tesla’s feedback to suppliers, Optimus still faces several hardware issues: overheating joint motors, low payload capacity for dexterous hands, short transmission lifespan, and limited battery life. Tesla is still testing samples from multiple dexterous hand suppliers, trying at least three different technical approaches. One Tesla insider stated that, currently, Optimus is only moving batteries in Tesla’s battery workshops, at less than half the efficiency of a human worker, and has yet to perform more complex vehicle assembly tasks. Tesla has been developing robot hardware with hundreds of supply chain partners for three years, creating a full humanoid robot supply chain. The Optimus R&D team exceeded 400 members by the end of last year. At a cost of around $60,000 per unit, Tesla's component expenses for Optimus this year alone were expected to surpass $300 million. None of the suppliers we spoke with had questioned Musk’s mass production plan—until now. "At the beginning, we were hesitant," said a senior exec from a Tesla supplier. "If you don’t believe in it but others do, the opportunity is theirs. You not only have to believe, but believe earlier than others." One of the earlier supply chain sources added that Tesla may debut Optimus Gen 3 at this year’s shareholder meeting.

English
8
1
85
14.5K