The interesting part isn't the humanoid, it's the job it got: the gap between two machines.
Sewing has been automated for years. Picking up a floppy collar and lining it up hasn't, because fabric changes shape the moment you touch it. Hence the tactile sensing.
What's the next floppy job factories can't hire for?
Humanoid robots join the garment factory workforce
The Aitu humanoid robot I shared earlier is now working at Bosideng’s(just like Canada Goose)intelligent factory in Changshu, Jiangsu, alongside sewing machines and other automated equipment.
Powered by a sewing-specific embodied AI model, vision and tactile sensing, it can separate, pick up and align fabric pieces before feeding them into machines for tasks involving collars, sleeves and front plackets.
The factory connects production equipment and tracks orders and quality in real time, with over 90% of key processes automated.
Yet flexible fabric handling still relies on skilled workers who are increasingly hard to recruit. Aitu takes on these tasks, while specialized machines handle standardized sewing.
The opportunity lies in automating the gaps between machines.
The most telling shot here isn't a robot, it's the wall of different bedrooms at the end. The first generation is an exposed frame on a lab floor, the middle one handles totes in a warehouse, the latest makes beds. That arc is really a change in the training data: a warehouse hands you the same tote a million times, a home hands you a new duvet, lamp and floor every time. The softer shell matters partly because it makes putting a robot in those rooms acceptable at all. How many rooms before a new bedroom stops being new to the policy?
Fonss retweeted
🚨 اختتام أسطوري لمؤتمر Nvidia GTC Berlin 2026: جينسن هوانغ يستعرض Hermes Agent في عرض خيالي يصعق الجميع!
بكلمات قليلة وبـ "برومبت واحد فقط"، قام جينسن بتثبيت وإعداد نظام Hermes مخصص، ليختتم به مؤتمره بضربة مدوية توضح كيف يُبنى مستقبل الروبوتات الذكية:
📌 تفاصيل العرض التقني المجنون:
وكيل يبني وكيلاً: الطريقة الأفضل لبناء Agent هي أن تطلب من Agent آخر أن يبنيه لك!
محاكاة روبوتية خارقة: باستخدام جهاز Nvidia RTX Spark، قام Hermes بالتخطيط لعائق تدريبي لبطة Hugging Face Microduck، ثم استعان بـ Nemotron لتخطيط المسار وبـ GPT-6 Astra لتنفيذه عبر FreeCAD، Omniverse، Blender، وIsaac Lab!
الواقع والمحاكاة: تدريب آلاف البطات الافتراضية على المسار حتى النضج التام، لينتقل الذكاء بسلاسة للروبوت الحقيقي في العالم المادي.
📌 الثورة الحقيقية للمصدر المفتوح:
المصدر المفتوح ليس مجرد فكرة فلسفية، بل هو أداة عملية تمنحك القدرة على ربط السحابة، النماذج المحلية، والأجهزة لتصنع سير عمل خارق بنفسك، دون انتظار شركة كبرى لتسمح لك بذلك!
احفظ التغريدة وشاركها مع أصدقائك المطورين.. المستقبل يُصنع الآن! 🧵👇
Returning the ball is the easy part for a tennis robot. The hard part is getting there: TRON 2 has to commit to a direction before the ball crosses the net, then keep correcting its guess mid-stride on wheels and legs at once.
That predict-then-correct footwork is exactly what a robot needs around people who don't move on schedule. Where does it show up first outside a court?
TENNIIX ULTRA MAX brings high-quality AI practice to everyday tennis. Enhanced Robotics × LimX Dynamics combine TRON 2’s all-terrain mobility with TENNIIX PRO: fast court coverage, jumps, precise net stops and real-time returns. Accelerating physical AI deployment.
#LimXDynamics #EnhancedRobotics #TennisRobot #EmbodiedAI
Fonss retweeted
Un détail : lors du show de présentation de la Tesla Roadster : le robot Optimus est au volant, il lève le pouce tout en retournant la voiture... puis éclate de joie 🤩
Fusées, voitures, robots, satellites, IA @elonmusk est partout !
Readers added context they thought people might want to know
La vidéo est fausse et a été générée par IA. La caméra d'Optimus ne correspond pas à la vraie scène et lors de la présentation officielle sur scène, on voit très clairement que c'est un mannequin avec une combinaison SpaceX qui était à l'intérieur du véhicule.
x.com/Optimus_RH/sta…
x.com/Tesla/status/2…
The flight is the headline, but the quieter story is parts reuse. The Roadster's active aero runs on the Optimus actuator system, and the hop comes from SpaceX cold-gas thrusters, not a new engine. Tesla is building cars out of its robot and rocket parts bins.
Car, or robot on wheels?
Track mode. Tesla says the wing actuation comes from the Optimus actuator system, so the same joint hardware moves a humanoid arm and a downforce wing.
xitter.cf/Tesla/status/211090513…
Roadster transforms in Track Mode
These aero features aren't just for looks: they provide the best passive downforce package of any road car (approaching F1 levels!) with wing actuation driven by the @Tesla_Optimus actuator system
The thrusters up close. Cold gas is the simplest rocket there is, which is probably why it made it into a road car.
xitter.cf/SawyerMerritt/status/2…
Fonss retweeted
Configurations
– Performance: road ready, seats 2
Active aerodynamics & multi-mode suction system for next-level performance on & off the track
– Thrust: race ready, seats 2
Multi-mode suction system paired w/ rear-mounted cold-gas thrusters that enable 0–60 in <1 second
– Rocket: air ready, seats 1
Multi-mode suction system plus perimeter-mounted cold-gas thrusters supporting forward & upward acceleration plus lateral rotational movement. So...yes, it can fly
Range >300 miles
Top speed up to >230mph (Rocket)
0-60
Performance (<1.6)
Thrust (<0.9)
Rocket (<0.59)
Reserve now –
Tesla.com/Roadster
Fonss retweeted
Roadster xitter.cf/i/broadcasts/1yKAPwNgg…
9-minute charging moves humanoid economics more than a better hand does. With 90 min of runtime, Digit spends about 9% of each cycle on the charger, so one robot can cover nearly two shifts without a spare unit standing by.
The catch is heat. Fast-charging every 90 minutes is hard on cells. Which gives first, pack life or that 20-hour target?
The interesting Gemini Argon demos aren't the prettiest ones. They're the ones where the model builds its own test bench first: a sim, a physics check, a game loop, then fixes itself against reality.
Which of these would survive contact with a real robot?
Nobody wrote a skill for Spot here. The model read the SDK docs itself, then built pick-up-can and press-the-pedal. Docs are becoming the robot's training data.
xitter.cf/i/status/2110057669172…
A full game world from one prompt. Same scene-building skill a robot needs to make its own practice sim.
xitter.cf/boldbug_dev/status/211…
The cornhole is the least interesting part. One human video became a self-written sim, and a few real throws corrected the physics it got wrong. That loop used to be weeks of an engineer's sysID work.
Does it still hold when the task is a grasp, not a toss?
Gemini 4 Argon taught a robot to play cornhole just by watching a video.
It mapped the physics, coded its own simulation, and taught @Apptronik's Apollo humanoid to sink 7 throws in a row.
Fonss retweeted
Esta skill te crea vídeos con HTML, CSS y JS.
Con animaciones y un editor integrado.
Se llama HyperFrames y es de código abierto.
Funciona en Claude, Cursor, Codex y Gemini.
→ github.com/heygen-com/hyperf…