Only straight to the point 🍟 Feel free to follow 🔁 and repost
Currently in Fukuoka, occasionally back in China, love tech, love robots, love selfies, love BTC
SANDO guarantees collision-free drone paths through unmapped spaces with moving obstacles, provided their maximum speed is known—and avoided all moving obstacles in 12 test flights. @MIT https://t.co/VaWSKjMpOr
1/3 MIMAS is out in Science Advances: a 2.31-g open-source microlens-array miniscope for cellular-resolution imaging across ~4 × 4.5 mm² of dorsal cortex in freely behaving mice. https://t.co/tCtj8thGT8
Robot data suppliers are everywhere now, however, it is the quality that determines whether your data is worth $200 or $2 per hour. Quality control can be very hard to do for large scale datasets, there are always imperfections like redundant motions.
NEEDLEWORK automatically identifies motion redundancies within a visuomotor dataset, then patches it with stitching using learned inverse dynamic models. The resulted dataset is cleaner, more efficient, has larger effective state coverages, and trains more successful policies.
So happy for @JuntaoRen to finally release the project! Effective dataset stitching has many practical difficulties and we have gone through a LOT of iterations for this to work, during which Juntao has consistently demonstrated professionalism and persistence, overcome one challenge after another. Congrats!
Check out @JuntaoRen 's posts for details!
GPU-Accelerated Path-Dependent Marginal Information Gain for Autonomous Exploration
João Félix Mendes, Rodrigo Ventura, Meysam Basiri
https://t.co/to3nSumpyC [𝚌𝚜.𝚁𝙾]
💬Submitted for review to IEEE ICRA 2027
The best robot training data may be the data that was never explicitly demonstrated!
Check out how NEEDLEWORK 🪡 stitches together suboptimal and failure trajectories to improve both demonstration quality and diversity.
Agentic Robotics -- using AI coding agents to design and fine-tune robot control systems using harnesses like ENPIRE, ASPIRE, or GaP -- are super exciting. If you know of any small or medium manufacturing companies who are applying Agentic Robotics to improve real production, please get in touch.
In many fields, automated research has become increasingly plausible with the improving capabilities of frontier language models, and the field robotics may be no exception. In ENPIRE, @_wenlixiao, @jiaxie_jason, and @TongheZhang01 build a harness framework that captures physical feedback and allows for automated robotics research, minimizing human effort while training policies that achieve a 99% success rate on challenging, dexterous tasks like organizing a pin box or fastening a zip tie.
Learn more in Episode 110 of RoboPapers, with @micoolcho, @chris_j_paxton and @ruijie_sg!
整个国庆节都在学习,几乎没怎么出远门(主要人也太多了),我真切的感受到, AI 留给普通人赚钱的机会越来越少了,以前是人与人之间的竞争,光在 24 小时卷就够了, 现在是人与芯片之间的协同,谁的 AI 调教的好,谁的脑洞够大谁就可以依靠 AI 获得成千上万倍的收益增长,而这一切都是源自于多年累积的对于事物判断的决策力。