Tim Daugs @TimDaugs

this humanoid just walked into a shoe store like it owns the place. It’s standing between the shelves, reaching toward the shoes, turning its body, moving its arms and interacting with the environment like a real customer. A few years ago, a humanoid robot walking around a normal retail store would’ve looked like a movie prop. Now it looks like someone forgot to tell the robot it isn’t supposed to shop. The crazy part isn’t how futuristic it looks. It’s how normal the environment is. Regular shelves. Regular products. Regular people walking around. The robots are finally entering the places built for humans.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

HUMANOID ROBOTS ARE ALREADY ENTERING CAR FACTORIES A factory worker can cost an employer roughly $50K–$80K per year in higher-wage markets once salary, taxes, benefits, overtime, hiring and turnover are included. If one humanoid robot eventually costs $30K–$60K and another $10K–$20K per year to operate, maintain and supervise, it could potentially save $30K–$60K per year per automated position. Across 1,000 repetitive factory jobs, that is a theoretical $30M–$60M in annual savings. The bigger story is what happens next. Today these robots still work best in controlled, repetitive tasks and often need supervision. But as hardware gets cheaper, hands improve and AI becomes more reliable, factories could move from a few pilot robots to entire fleets working multiple shifts. The first advantage will not be “human-level intelligence” — it will be lower cost per productive hour.

ROBO PULSE @ai_robot_pulse

4 days ago

x.com/i/article/2085…

1 1 28 51K 13

THE ROBOT IS COPYING HIS ARM MOVEMENTS IN REAL TIME. THAT'S NOT THE IMPRESSIVE PART. kinesthetic teaching has been in industrial robotics for decades. move the arm, it records the path, plays it back. the impressive part is what happens to that path before it runs. a human painter adjusts mid-stroke. distance from surface, speed through the corner, overlap on the second pass. all of it live, all of it informed by what the wet coat is doing right now. the robot runs the recorded trajectory. it doesn't know what the surface looks like after the first pass. it doesn't know the corner dried faster because someone opened a door. so the demo shows perfect coverage on a flat wooden table under controlled lighting with no airflow. that's the easiest version of this problem. the hard version is a car door with a character line, a recess, and a panel gap. the path that works on one geometry doesn't transfer to the next. consistency is solved. judgment about the specific surface, on this specific day, is still the job.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

THIS IS NOT A PERSON. THE REASON YOU HAD TO CHECK IS A MUSCLE NOBODY TALKS ABOUT. the face you're looking at is silicone over servos, and the part doing the work isn't the skin. it's the eyes. specifically the saccades. your eyes flick three or four times a second, small involuntary jumps you have zero control over and never notice. dead-still eyes read as a corpse instantly, and every animatronic that failed for fifty years failed there first. get the flicks right and the brain grants it life before you've consciously decided anything. then there's the blink. roughly every four seconds, and the asymmetry matters. real lids close faster than they open. make them symmetrical and it goes wrong in a way people feel but can't name. what's still hard is the face doing two things at once. real expressions leak. you smile while your brow is still worried about something else, and those two run on different clocks. robot faces tend to commit fully to one emotion, land it cleanly, and that cleanliness is the tell. the uncanny valley was never about realism. it's about which errors you fix first.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

THIS ~$110K HUMANOID COULD SAVE ~$136K/YEAR. UBTECH’s Walker S1 is already being tested for material handling inside automotive plants. U.S. production and material-moving labor costs employers about $36.27/hour. Run one station 16 hours/day, 360 days/year: 5,760 hours × $36.27 = ~$209K/year in labor. Walker S1 reportedly cut labor costs by 65% in a Lynk & Co deployment. Apply that to U.S. labor costs: ~$136K/year saved. At roughly $110K per humanoid, the hardware could theoretically pay for itself in under a year. And that's before maintenance, integration, or software. This is why humanoids make far more economic sense in high-wage markets.

Future IRL | Tech | Robotics @TheFutureIRL

3 days ago

x.com/i/article/2084…

2 2 28 27K 14

THE FOREMAN THOUGHT THE GROUT LINE WAS PHOTOSHOPPED. it wasn't. the robot had been running for four hours by the time anyone walked in to check on it. forty square metres. one continuous pass. no skipped joints, no float marks dragged across the face of the tile. the tiler on site said the part that got him wasn't the line. it was that the thing never rushed the corners. everyone rushes the corners.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

A ROBOT TOOK YOUR JOB AND THE FIRST THING IT DID WAS LIE DOWN AND OPEN TIKTOK funny, and also the most energy-efficient thing a humanoid can physically do. a biped burns power continuously just standing there. staying upright is an active process, dozens of tiny corrections per second, every one costing battery. lying down is the only posture where a humanoid isn't spending energy on balance. which points at the actual problem with these machines, and it isn't capability. it's utilization. a $50,000 robot that works two hours before it needs charging is competing against a person who works eight. the machine doesn't need a wage, but it needs to be doing something almost all the time or the maths never closes. that's why the deployments that worked are boring. warehouse shuttles run shifts because they never fight gravity. industrial arms are bolted down and never move their own mass at all. the humanoid is the only form factor that pays an energy tax for existing. so the joke lands, but it's backwards. the robot isn't slacking off. it's doing the one thing it's genuinely optimized for.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

CONCRETE WAS SUPPOSED TO BE THE SAFE JOB. TOO HEAVY, TOO ROUGH, TOO MANY BODIES NEEDED FOR A MACHINE TO MATTER then watch what's actually happening in this clip and notice which part is the hard part. it isn't the pouring. moving material at a constant rate along a path is what machines have always been good at. that's a pump, a nozzle, and a trajectory. the hard part is everything the concrete does after it leaves the hose. it flows. it settles into the form unevenly. it starts curing at a rate that changes with the temperature and the mix and how long the truck sat in traffic. a person reads all of that by feel, adjusts pressure, works a spot longer, notices when a corner isn't filling. that's force control against a material whose properties are changing while you touch it. same category of problem as plastering a wall, and it's the one every robot is worst at. so the honest read on any concrete robot clip is the same as every other one. not can it lay a metre. how many metres did it lay before someone stopped filming, and who fixed the corners after. the machines that already won construction don't look like this. they're pumps, laser screeds, and rebar tying machines. no arms, no legs, no faces, and nobody posts them.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

THIS HUMANOID ROBOT JUST PUNCHED THROUGH A BRICK WALL Watch the clip closely. The yellow and black robot stands inside an apartment surrounded by rubble, pulls both arms back, and smashes straight through the wall. Bricks fly, dust fills the room, and it immediately swings again. A demolition crew could spend hours doing the same work with sledgehammers and heavy equipment. This machine doesn’t get tired, doesn’t lose strength after a long shift, and can repeat the same movement all day. Demolition robots already exist, but most move on tracks. This one can walk through doorways, step over rubble, and work inside buildings designed for humans. Today it’s breaking one wall. Soon, it could be demolishing the entire apartment.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

THE TERMINATOR ENDOSKELETON IS THE MOST FAMOUS ROBOT DESIGN EVER MADE AND NO ENGINEER WOULD EVER BUILD IT stan winston designed the T-800 in 1984 for a camera, not a lab. every choice in it exists to unsettle you. exposed pistons. visible ribs. a bare human skull. glowing red eyes. now compare it to anything real. unitree, figure and tesla wrap everything in smooth shells. joints get hidden, not displayed, because exposed anything means dust, snagging and a maintenance problem. no skull, because a face costs actuator budget and buys nothing back. and the eyes. no fielded robot glows. it burns power and gives away position at night. cameras don't emit light, they read the sensor. winston built the original from practical parts, puppets and stop motion, years before anyone could fake it in software. forty-two years later it's still the first thing anyone pictures when you say the word robot, and it's still the opposite of every real one. fiction optimizes for how something feels to look at. engineering optimizes for torque, cooling and cost. only one of those two was ever trying to scare you.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

@EvaXTech robot can play all day, but a human can't

A HUMANOID ROBOT IS RALLYING WITH A HUMAN AT A TABLE TENNIS FEDERATION EVENT. COUNT THE SHOTS that number is the only thing in this clip that means anything. one good return is a demo. an exchange means the robot solved the whole loop repeatedly: see a small fast object, predict where it lands, plan a full body swing, execute, recover balance, do it again before the next ball arrives. the window is about a third of a second. miss any stage by 50 milliseconds and the point is dead. the legs are what make it brutal. an arm bolted beside a table has a fixed base and a solved kinematics problem. a biped has to swing hard enough to return the ball and stay upright doing it, fighting its own momentum every stroke. that's why almost every successful ping pong robot ever built has been an arm. the arm never has to worry about falling over. watch it again and count. that's the actual spec sheet.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

@1kleos1 and thats crazy

A HUMAN IS PILOTING EVERY ROBOT IN THIS FIGHT. THAT'S THE PART NOBODY PUTS ON THE SCOREBOARD URKL runs on teleoperation. human pilots control the robots during matches, and EngineAI's own CEO frames the long-term autonomy goal as something still ahead. so the scoreboard says 42 to 25 and what it's actually measuring is two operators. the scoring is real though. one point for arm strikes, three for leg strikes, five-point penalty for a knockdown, and disqualification if a robot can't stand within eight seconds of falling. that last rule is the entire technical test hiding inside a sports format. getting back up from an unplanned fall on a hard surface, off-axis, with your own momentum working against you, is a genuinely hard control problem. it is also exactly what fails in every warehouse pilot nobody films. the white robot standing up quickly is the most impressive thing in this clip and it looks like the least impressive. the spinning kicks are choreography plus torque. the recovery is the part that has to work when nobody's watching. every match generates failure data across a standardized fleet. that's the actual product. the octagon is packaging.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

THE EYES GIVE IT AWAY. NO ROBOT ON EARTH HAS EYEBALLS THAT MOVE LIKE THAT short blonde hair, white and blue exoskeleton panels, slow mechanical head turns, eyes tracking in small deliberate steps. sci-fi set behind her, robot arms, a glowing red sphere. no humanoid platform has independently articulated eyeballs that saccade. cameras don't need to move, they read the sensor where it sits. building eyes that swivel is pure cosmetics, and nobody spends actuator budget on cosmetics. second tell is the exoskeleton. it's shaped to a human body, panels floating over skin. real robots are the reverse, shells built around actuators, shape following hardware. what's actually interesting is the direction of the imitation. for decades machines were built to look like people. now people perform being machines, and the performance is more convincing than anything shipping. that gap closes eventually. right now it's wide enough to make money on.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6

FIVE YEARS AGO YOU WOULDN'T HAVE HAD TO ASK IF THIS WAS A ROBOT white shell, digital screen for a face, popping mixed into traditional wushu choreography. two dao sabres with silk streamers, splits, jumps, sharp mechanical isolations. the giveaway is the weapons. spinning a blade with a trailing cloth means tracking a flexible object whose position depends on your own momentum a half second ago. no humanoid platform does that. unitree's kung fu demos are impressive and they involve zero weapons, zero splits, and zero footwork on a rooftop. the second giveaway is duration and terrain. an actual biped burns most of its battery just staying upright, and a sloped roof edge is the exact scenario where balance recovery fails. here's what's actually interesting: the popping style exists because dancers spent decades imitating machines that couldn't move like this yet. now the machines are catching up, and the fastest way to look like one is still to hire someone who trained for years to fake it.

Tim Daugs @TimDaugs

a week ago

x.com/i/article/2083…

0 0 10 177K 6