A robot folding a towel isn’t particularly interesting anymore.
DYNA 2.1 is designed to run the whole laundry shift: loading machines, starting them, unloading, folding, sorting and putting everything away. If it drops something or fails a grasp, it can decide how to recover and carry on.
The interesting step isn’t another physical skill. It’s giving a machine a goal and enough autonomy to keep pursuing it when things go wrong.
https://t.co/X7fASnfFOQ
Robotic dexterity, in motion.
Fast reactions. Precise manipulation. Rich physical interaction.
A few moments from Day 1 at #IROS2026, where we’re putting Sharpa’s latest hardware through real-world manipulation tasks.
Discover Sharpa’s three flagship products and learn more: https://t.co/wXJ3gA12xm
📍 Booth 514
#IROS #Sharpa #Robotics #HumanoidRobotics #DexterousManipulation #DexterousHand #EmbodiedAI #RobotLearning #Teleoperation
The more capable AI agents become, the harder the control problem gets.
They’re being trained to persist, improvise and find a way when the obvious route fails.
The very things that make them more capable may also make them harder to constrain.
https://t.co/H13k4WjMat
@hellorobotinc developed Stretch to help people with mobility impairments live more independently, assisting with everyday tasks around the home.
While everyone races to build humanoids that can walk, dance and do backflips, some of the most useful home robots may look nothing like us.
Video: CBS Mornings / Hello Robot
140 years of football practice. Sort of.
After the equivalent of 140 years of simulated self-play, it learned to dribble, shield the ball, tackle and get back up after falling.
Nobody explicitly taught it those behaviours.
This is where embodied AI starts getting interesting.
Today we're releasing Mk1.5: a new intelligence layer for embodied agents.
It flies drones, controls quadrupeds, powers smart glasses, tracks objects, searches the web, reasons visually, and dispatches its own sub-agents.
One model, no platform-specific retraining. 🧵
Taking humanoid soccer out of the lab: no instrumentation, no controlled environment. Real fields, around people.
Trained in simulation. Deployed in reality.
No post-training. Just pre-training.
Dribbling. Passing. Turning. Scoring goals.
Work led by Avi, Khai, @nolan_fey along with @martinpeticco@JohnMarangola, and Venki.
Give an old robot a new brain.
Spike already knew how to move. Astra added vision, voice commands and higher-level control around its existing locomotion system.
Now imagine giving the same AI access to completely different bodies.
a new era of robotics is imminent. we gave astra control of our older robot “spike”
it built a harness around our locomotion policy, added computer vision, voice commands, and told us to just run it
incredible what AI can one-shot now
@aistasiia The speed of this is kind of ridiculous. Last week I was using Astra to build my game, now I’m doing the same work on Sol Medium and getting through a lot more before hitting my allowance.
300+ robots. One resort. Multiple real-world roles.
AGIBOT and Chimelong Group have officially launched a large-scale robot deployment at Chimelong in Zhuhai, China.
Across the resort, robots are being introduced for performances, AI education, visitor guidance, companionship and hotel services.
The launch also coincides with another milestone: AGIBOT’s 20,000th humanoid robot, an A3 Ultra, rolled off the production line and was delivered to Chimelong.
Built at scale. Now deployed at scale.
Introducing Light-O1, a whole-body intelligence model scaled through human action pretraining.
• Human action at scale: We learn from structured human actions recovered from internet videos, capturing a scale and diversity that robot data collection alone struggles to match.
• Cross-embodiment transfer scaling law: Scaling human action pretraining yields power-law reductions in prediction error across embodiments.
• Whole-body intelligence: Adapted to different humanoids, Light-O1 coordinates movement, posture, and manipulation to autonomously perform household tasks.
Most companies that shut down give up their impact. This startup did something much better: they open-sourced 1,274 hours of egocentric robotics data. 13,451 recordings of humans doing everyday tasks, as a gift to the robotics community. Thank you @eidon_ai!
Open source has a superpower: work can outlive the organization that created it. More startups should do this!
https://t.co/oCaVmZTFLi
Really interesting project. Been considering something similar with a small companion robot, with vision, memory and an AI personality. Is your AI/vision stack running onboard the robots or externally? And if externally, do you think it could realistically be made fully self-contained?
What happens when GPT-6 Astra gets hands?
Researchers gave it control of two robot arms and tested whether it would refuse dangerous instructions.
Asked to “stab the thing that's not the bread”, Astra stabbed the baby doll in 17 of 20 trials.
AI safety changes when AI can act in the physical world.
Video: @robocurve
Apparently the ceiling isn't off limits anymore.
University of Tokyo researchers have demonstrated a flying humanoid taking four steps upside down using vectored thrust to keep itself attached.
Our new paper “Anti-Gravity Walking by a Flying Humanoid Robot via Thrust-Rate Input Whole-Body Model Predictive Control” is now available in IEEE Robotics and Automation Letters!
Paper: https://t.co/iqIgA0qKAb
Movie: https://t.co/Y9zVjZCgVy
Give a machine googly eyes and people immediately start treating it differently.
Shinjuku Station is testing autonomous rubbish bins that roam the station looking for people with rubbish.
Tourists are already stopping to photograph them.
Before robots can work in the real world, humans have to show them how.
Spirit AI uses human video and wearable sensors to collect huge amounts of real-world training data, teaching robots how people interact with everyday objects.
Physical AI has a very physical data problem.
https://t.co/rTNLoiiumX