@micoolcho
I ❤️ robots, cheap hardware, steam engines, XGBoost, Liverpool FC & SG 🇸🇬 | Plane crash survivor | Building @BitRobotNetwork @frodobots
Archon Robotics @Archon_Robotics🤖 Building Whole-Body Intelligence for Tactile Sensing and Deformable Object Manipulation. Touch supports airpods grasping; coordinated hands handle changes in fabric shape.
Siyuan Huang @siyuanhuang95Do we really need so much real robot data to train a good whole-body loco-manipulation model? Can we use GPT-Astra to boost sim2real capability? Excited to share Zero-HLM: Generalizable Humanoid Loco-Manipulation without Real Robot Data! 🤖 Our core insight: use real2sim2real to develop humanoid loco-manipulation skills that generalize—without collecting real-robot demonstrations. 📱 Capture real scenes with a smartphone 🌐 Build visually faithful, physics-ready digital twins with Astra 🦾 Generate whole-body demonstrations in simulation 🚀 Transfer learned policies directly to a real humanoid With simulation-only task training, policies transfer zero-shot to a Unitree G1 across three long-horizon tasks. Across five generalization settings, pooled real-world success rises from 23% with real-data training to 69% with our simulation data. Ground simulation in reality, scale experience in simulation, and bring the skills back to the real world. Proud of the team behind this work! 🔗 Project: and paper:
Qiayuan Liao @qiayuanliaoSuperMimic on musculoskeletal humanoid, a much more complex embodiment!
Chenhao Li @breadli428Compared with the speed gain from “neural-network-izing” dynamics, I think the training speed improvement from using first-order gradient with BPTT over zero-order policy gradient is minor. This gain is even more ignorable with the learned WM, where samples are incredibly cheap.
Reward AI @RewardAI_We deploy OM-1 on more humanoid tasks in both indoor and outdoor environments. OM-1 can power humanoids to do useful things robustly that other models or form factors cannot. - last-mile delivery from a Cybertruck - manipulation on stairs
Shane @shinheeshaneleeOur @DeftRobo robots have helped build ~20,000 Tesla Model 3/Ys. Bought one recently? Our robots likely handled its parts. Production-deployed at a Tier 1, on robot foundation models + our deployment infra: - 700 parts/day - 24 hrs/day - 100% uptime Next: triple-digit robots
Edward Johns @Ed__JohnsToday we’re out of stealth, and we’ve raised a $40m seed round co-led by @creandum and @usv, to bring Instant Robot Learning to life. For the past few months, we‘ve been busy building a new company. Machenta. This follows results from a new kind of robot foundation model we’ve built, which I’m so excited to share with you today! Usually, it takes hours/days of data collection and training to teach a robot. We’ve built a model that learns tasks instantly: one demonstration, no further training. A model trained entirely in simulation, without a single pixel of real-world data. A model where teaching robots is so quick and easy, it feels like physical vibe coding. A model for robots that learn like humans do, from someone like you. Welcome to the era of Instant Robot Learning. Thank you @Kathi_Wilhelm, @rebeccakaden, @hkanji, @jonno_evans and all of our angels for your support! @Machenta_ Learn more: See below for what you can do with Instant Robot Learning!
a16z @a16zComputer use is becoming faster than humans, and cheaper too OpenAI's Ari Weinstein: "We've gotten computer use now to a place where it's faster than the average human at using a computer, and I think we're getting really close to a place where it is superhuman." And it already costs a fraction of human labor per hour of back-office work: - Computer-use agent: $6-8 - Offshore outsourced talent: ~$10 - US back-office labor: $30-45
Praveen Kumar Verma @Alacritic_SuperPurdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks. By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time. Massive implications for next-gen search & rescue and edge AI.
Rishabh Karnawat @rishabhkioMany of you don't understand how serious this is, so let me break it down. In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force. For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew. About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea. They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science. Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption. Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean. And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea. So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited. Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved. But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.
铁锤人 @lxfater卧槽，3090要可能要涨价了！！ 因为老黄的刀法被人破解了 有人给3090解了个锁，同样限制在 360W，PyTorch 测试从 75 TFLOPS 跑到了 135 TFLOPS，提升约 80%！ 你以为是超频，把功耗往上拉？ 但他改的是显卡里控制运算速度的寄存器，这些寄存器平时有权限保护，正常改不了 他利用 GPU 启动固件的漏洞，赶在 NVIDIA 驱动加载之前拿到写入权限，把相关运算改成全速，再交给原版驱动使用 解开的主要是 FP16／BF16 输入、FP32 累加这类 AI 运算的速度限制 但测试算力多了 80%！！ 4090 也有分支项目做到了：同类运算的指令吞吐，从约 165 TFLOPS 提升到 331 TFLOPS，接近翻倍 项目地址，附上👇
Abdul Rahim Mirani @rahim_miranieveryone's fighting over robot data and i think most of it is the wrong kind teleop gives you clean demos of the easy 90%. the hard part is the failures and recoveries, and nobody logs those because the operator just retries @physical_int showed this with pi*0.6 on a real espresso task. they took the model trained on normal demos and put it on the robot. then they added about 400 human corrections (someone stepping in when it messed up) plus its own runs, failures included, and trained on all of it throughput more than doubled and success went past 90%. same model, same robot, just a different kind of data. they say it roughly halves the failure rate on the hardest tasks too the company that figures out how to collect failure data cheaply wins
Kyle🤖🚀🦭 @KyleMorgensteinthe pace of research now is wild. supermimic came out 3 days ago. no code release. now there are already derivative works with full code. it’s hard not to feel the pressure to always move faster.
Jon Barron @jon_barronThis bug was first found and fixed by Gemini Flash 3.7, and most experiments in the paper were done by Flash 3.8. Gemini 4 Argon wrote the first draft of the paper and figures, which I aggressively nitpicked until the content was solid but the writing was not. Then I fed Argon a handful of my first-author papers from the last few years and had it do a rewrite in my style (which worked astonishingly well), and then I manually did a rewrite (the hands + keyboard variety) after that. I really liked this workflow! If you like my writing style feel free to throw some of my old papers in your context window and do the same.
Toru @ToruO_OHow should we approach research, when the very field we work on is going through rapid and potentially fundamental changes? I recently gave a talk at the Stanford Vision Seminar, reflecting on my PhD journey and how AI is changing the way I think about research. I adapted part of it into a new blog post, organized around three questions: 1. What should we work on? 2. How should we work with AI? 3. What is a PhD for? I hope some of my reflections are helpful. The talk ended with an open discussion, and I’d love to continue it here; curious how others are thinking about these questions. Personally -- as much as there's FOMO, I also think life is very long, and we have time.
RoboPapers @RoboPapersSpecifying reward functions for robots is one of the hardest things about reinforcement learning. Robot rewards often need to be very detailed; metrics like progress can be ill-defined and hard to estimate. This leads to most robot learning defaulting to sparse rewards or simple preference learning. But naive preference learning (having human annotators choose one trajectory over another) is an easy solution, but obscures a lot of the signal in complex tasks and can make learning a lot less efficient. @marceltornev, @anubhamahajan01, and @AbhijnyaBhat join us to talk about their solution: freeform preference learning, which lets annotators define natural-language axes to compare trajectories over. This improves real-world performance on long-horizon manipulation tasks over sparse rewards and simple binary preference learning. Watch Epsiode 103 of RoboPapers, with @micoolcho and @DJiafei, today to learn more!
Jeremy Collins @jerthesquare_This made me question: do we need fine-tuning at all? It improves performance for 10 of our 12 beta tester robots. Fine-tuning also of course helps to specialize Dream-1 to specific tasks. However, onboarding a new robot often just doesn't require fine-tuning anymore.
C. Zhang @ChongZzZhangWith @JayHe748646 @KehanWen170077 @leggedrobotics , we push humanoid mobility to next frontier. Achieved by a generalist contact planner + a generalist tracker + a generalized neural mapping, warmstarted from few human data. Website: Promo Video 👇
