Ghostin

AI researcher and creator

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

Humanoid robot stands over a conveyor belt buried in parcels - soft mailers, boxes, plastic-wrapped bags - picks one up, checks the shipping label, and sets it into a sorting lane. Then the next one. Then the next.

This is the least flashy kind of robot footage possible, and that's exactly why it's worth more attention than a kung-fu routine. Per the humanoid robot race breakdown, the one company actually winning right now isn't the one with the best hardware specs or the biggest funding round - it's Agility's Digit, which has quietly moved over 100,000 warehouse totes under real paying contracts with Toyota and Mercado Libre. No viral clip. Just totes, every day.

Parcel sorting is the same category of task: repetitive, structured, physically simple but tedious enough that human turnover in these roles is brutal. It's exactly the kind of job that doesn't need a robot that can improvise - it needs one that shows up for every single shift and never asks for a break. That's a boring pitch. It's also the only kind of robot pitch with real receipts behind it so far.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

@beamnxw actually useful method you use for yourself

[Ghostin @ghostinlatent]

@shevaxgod sounds like a conspiracy to me

[Ghostin @ghostinlatent]

@thegreatest_sv that is a very niche way to use Claude

[Ghostin @ghostinlatent]

@0xTenzo very useful machine, is there any other robots that can work at construction?

[Ghostin @ghostinlatent]

Agility's Digit moved 100,000 totes, but this humanoid is already sorting live parcels on a warehouse line - grabbing poly mailers with both hands, scanning labels, and routing black, blue, white, and yellow packages into separate piles without ever taking a break.

It stands at a metal sorting station next to a conveyor belt, methodically picking up each parcel, examining it, and placing it exactly where it needs to go, handling soft bags and rigid boxes with the same steady grip. This is not a demo mat or a staged factory tour; this is the messy, high-volume reality of last-mile logistics where every second counts.

While Tesla Optimus is still "primarily for learning" and Figure is ramping toward one robot an hour, machines like this are already proving that the real money in humanoid robotics is not viral kicks or earnings-call hype - it is doing the boring, repetitive, physically exhausting work that humans do not want to do at 3 a.m.

The physical AI race is not about who walks the smoothest; it is about who can survive a real shift, and this one just clocked in.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

Robot rolls a massive tractor tire across a yard on its own weight, tips it upright, and shoves it up onto the bed of a flatbed truck - solo, no assistance, no forklift.

Worth the reality check: per the robot race breakdown, even Atlas - the strongest published industrial hardware in the entire field, 56 degrees of freedom, IP67-rated, autonomous battery swaps - is rated for a 50 kg instantaneous payload. A tractor tire like this typically weighs 90-230 kg. Nothing currently deployed solo-lifts something this size the way the clip shows.

This is exactly the "physical AI" gap the article gets into: every company on that list can already build a body that walks, balances, and grips something. None of them have the brain-and-body combination yet to reliably improvise raw-strength tasks like this outside a controlled setting. The dexterity and force-judgment problem - not battery life, not joint count - is the actual bottleneck the entire $40 trillion bet is racing to solve.

The honest version, happening right now instead of in a render: Agility's Digit has moved 100,000+ warehouse totes with real paying contracts. Figure loads sheet metal on BMW's line. Smaller loads, verified numbers, no tractor tires yet - that gap is the nearest future the article maps out, not the one this video is already claiming.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

Every robotics pitch deck this year leads with the same number: $2 an hour. That's what a humanoid robot costs to run - cheaper than almost any human wage on earth.

Here's what that number quietly excludes: the purchase price. Add it back - a $50K robot, integration, task-specific programming - and the real effective cost lands at $3-10/hour, not $2. Still a fraction of $30-45/hr warehouse labor or $35-55/hr manufacturing labor, fully loaded.

The one real case study with verified numbers: BMW's Figure robots on the Spartanburg line, $130K per unit, offsetting $50-70/hr labor, paying back in under 2 years - already built 30,000+ vehicles.

The catch nobody puts in the pitch: payback swings from 6 months to 15 months on utilization alone, same robot, same task. Most year-one deployments run 45-70% utilization, not the 24/7 the marketing slide assumes.
https://t.co/TdfK73XylV

[FLY 🗣 @FLYingBanger]

https://t.co/trW6rCXFkV

[Ghostin @ghostinlatent]

@poetiqha soon they will replace us😅

[Ghostin @ghostinlatent]

Robots line up at the start of a 100m sprint on a real Olympic track, race down the lane at a jerky, mechanical run - then later, on a football pitch, one collapses face-first mid-match while staff rush in to check on it. This is real footage, not staged: the first-ever World Humanoid Robot Games, Beijing, August 2025. The scale is bigger than most coverage let on: 280 teams, 16 countries, over 500 robots, 26 events - everything from the 100m and 1500m to football, kickboxing, dancing, and even practical tasks like sorting medicine and cleaning.

Unitree - the same company that IPO'd this week for $9 billion, per the robot race breakdown - swept gold in the 1500m, 400m, hurdles, and relay. A robot called Tiangong Ultra won the 100m outright in 21.5 seconds. Tsinghua University's team beat a German squad 1-0 in the football final.

One observer at the Games noted something a human athlete would never do: when one robot fell in the relay, the ones near it went down too - no instinct to catch themselves independently, because that instinct hasn't been trained yet. Beijing has already confirmed a second edition for August 2026 - this isn't a one-off stunt, it's becoming an annual scoreboard for the entire industry, the same way early car races once were for the auto industry.

https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

@FLYingBanger actually interesting thought about robot perspective, I wonder if they can replace all of our work places

[Ghostin @ghostinlatent]

@Yumzlef glad this helped you

[Ghostin @ghostinlatent]

@0xKarmi they look almost the same

[Ghostin @ghostinlatent]

Tesla says Optimus is "primarily for learning," and Agility's Digit may have moved 100,000 warehouse totes, but this humanoid robot is already outperforming both by doing real construction work - it is plastering wet cement on a rough wall, scooping mortar from the floor, and smoothing it across the surface with a trowel in precise, repetitive strokes.

The article makes it clear that China controls 85–90% of humanoid volume, and while Western robots are still proving themselves in structured warehouses, this machine is handling gritty, unstructured physical labor on an actual job site.

This is physical AI in its rawest form - not a viral demo or a factory line, but a dust-covered robot mastering the messy reality of construction work. The race was never about who walks the best; it is about who can survive a real shift, and this one just clocked in.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

Tesla says Optimus is "primarily for learning," and Agility's Digit may have moved 100,000 warehouse totes, but this humanoid robot is already outperforming both by doing real construction work - it is plastering wet cement on a rough wall, scooping mortar from the floor, and smoothing it across the surface with a trowel in precise, repetitive strokes.

The article makes it clear that China controls 85–90% of humanoid volume, and while Western robots are still proving themselves in structured warehouses, this machine is handling gritty, unstructured physical labor on an actual job site.

This is physical AI in its rawest form - not a viral demo or a factory line, but a dust-covered robot mastering the messy reality of construction work. The race was never about who walks the best; it is about who can survive a real shift, and this one just clocked in.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

Optimus sweeps a floor with a brush, ties off a full trash bag and drops it in a bin, then stands swaying gently in place for what the caption calls "all day."

Then the same clip shows you the trick: split-screen, first-person camera views - a human hand in a glove, wrist-mounted camera, physically wiping a counter, rinsing a sink, dusting a shelf. The caption says it outright: teleoperation.

A person, wearing the robot like a puppet from a control rig, was doing the actual work the whole time footage like this gets filmed. That's not a scandal - Tesla has said openly that today's Optimus units exist mainly to collect training data for a future autonomous version, not to work unsupervised yet. But it's the exact gap the rest of this year's coverage keeps circling: Figure has BMW running its robots on a real line, Agility's Digit has moved 100,000+ warehouse totes with no human in the loop. Optimus, as of this clip, still has a person's hand inside the glove.

The sweeping and the trash bag aren't fake. The "on its own" part is what's still being built.
https://t.co/1CfVLQ14kA

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

А humanoid robot stands in a flooded rice paddy, bends down, plants seedlings by hand - grab, push into mud, straighten, step, repeat - row by row like a human transplanter.

The twist: rice transplanting was already solved over a century ago, and none of those machines look human. 98% of Japan's rice fields are planted by riding transplanters, some now fully autonomous with GPS and computer vision, no legs required.

Same theme as the whole humanoid race: the robots actually shipping work - Agility's Digit moving 100,000+ totes, Kubota's autonomous transplanters - win by not looking like us. Human shape is for spaces built for people. A rice paddy never was one.

The real question isn't "can a robot plant rice" - answered since 1965. It's whether general-purpose humanoids will ever beat purpose-built machines already doing it cheaper.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

@NEXORAResearch for this situation we have robots

[Ghostin @ghostinlatent]

Humanoid robot vacuums a living room floor, then walks to the kitchen, wipes down a table, and carries a cup of tea over to an elderly woman sitting at breakfast - setting it down gently in front of her.

Worth saying upfront: this looks like a concept/promo render, not a verified commercial product - the smooth, seamless motion isn't matched by anything currently shipping.

China's GigaAI has 100 actual home robot units placed this year in Wuhan households that have elderly members, children, or pets - explicitly testing this exact scenario, not a lab demo. The target: under $14,700 per unit by mid-2027.

The gap between the video and reality isn't the walking or the tea-carrying. It's trust: putting a robot alone in a room with someone's grandmother, unsupervised, for anything beyond carrying and tidying, is still years out.
https://t.co/1CfVLQ0wv2

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu

[Ghostin @ghostinlatent]

Today, the company behind those viral kung-fu robot videos just became a $9 billion public company.

Unitree priced its Shanghai IPO this morning - mainland China's first publicly listed humanoid robot maker. Revenue quadrupled to 1.7 billion yuan in 2025. One of its strategic investors: DeepSeek.

The backstory makes it land differently. Before this was a $9B company, Unitree was the one putting robots through kung fu routines and dance battles at China's Spring Festival Gala - a show watched by over 600 million people, three years running. That "just for views" reputation is exactly what funded the actual engineering: a $13,500 humanoid you can order today, running on the same tech that's now beating human half-marathon world records in open competition.

The catch sitting under all of it: only 13.3% of Unitree's revenue comes from the US, and new restrictions on foreign humanoid robots could shrink that further. The kung fu videos got you to notice China's robot company.
https://t.co/1CfVLQ14kA

[Ghostin @ghostinlatent]

https://t.co/whb6WHSYAu