larwick

researching AI, humanoids, drones & autonomous machines

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

The best explanation of how a robot boat sailed straight into a Category 4 hurricane. 120 mph winds, 50-foot waves, and a 7-meter drone called Saildrone filmed it from the inside.

Its sail is a rigid wing, and the push it gets from the wind grows with speed squared: lift = ½ρv²·A·C_L. Nine times the wind speed means 81 times the force, so the same wing that drifts along in a breeze has to survive a hurricane.

5 minutes. bookmark & watch today. one formula decides whether the wing flies or snaps

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

Finally, someone explained why Amazon's drones were set to lose $63 on every package. That number came from Amazon's own internal projections for 2025, reported by Business Insider.

A hovering drone holds its whole weight up on nothing but air, and helicopter physics says the power that takes grows with weight to the 3/2 power and shrinks with bigger rotors. That's why doubling the battery doesn't double the flight time: the extra battery is extra weight you also have to hold up.

11 minutes. bookmark & watch today. a $63 loss to fly a toothbrush to your yard

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

This is the simplest explanation of how a Roomba cleaned your room for years without knowing where it was. More than 50 million were sold, and the famous ones had no map at all

It hit a wall, turned at a random angle and drove on, and random lines still cover a floor if you draw enough of them. Drive about 4.5 times the floor's area and 99% of it is clean, whether it's a closet or a ballroom

5 minutes. bookmark & watch today. a robot doesn't need to be smart to finish the job, just patient

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

The most satisfying explanation of robot running I've ever seen. A pair of robot legs with no head and no arms ran 100 meters in 24.73 seconds, set a Guinness World Record, and the company behind them agreed to go public at about $2.5 billion

Running is just bouncing: your leg works like a pogo stick spring, catching your weight and throwing it back up, and a stiffer spring means shorter steps and a faster stride. Motors make bad springs, so the robot used real springs in its legs and learned its rhythm in a computer simulation first

10 minutes. bookmark & watch today. one pogo stick formula decides how fast any legs can run

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

This is how robots find a plane on the ocean floor, explained better than anywhere else. Governments spent up to $200 million searching for MH370 and found nothing. Now a robot company gets $70 million only if its machines find it.

Every empty search isn’t a waste - it makes the places you haven’t checked yet more likely, so the map of where to look keeps shifting.
Trust your sonar too much and you walk right past the answer, which is exactly how one lost jet stayed hidden for two years, until robots went back to the middle and found it in one week.

11 minutes. Bookmark & watch today. The one rule every search robot uses to decide where to look next.

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

Every drone, every helicopter and every flying car pays the same tax just to stay in the air: hover power grows with thrust to the power three halves. Double the weight and you don’t need twice the power. You need 2.8 times.

Momentum theory, worked out in the 1800s for ship propellers, says a rotor holds itself up by throwing air down. Bigger propellers move more air more slowly and save power. Thin air at altitude costs more. A 249-gram camera drone needs only about 9 watts in theory, really draws closer to 38, and that’s why its 19 watt-hour battery buys about 30 minutes. Double its weight with the same propellers and the hover time drops to about 15. Neural network pilots now outfly human champions, but no algorithm escapes this exponent.

11 minutes. Bookmark & watch today. The exponent behind every drone’s flight time.

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

A bike with nobody on it falls over the moment you let go. A robot motorcycle stays up with one line: the tangent of the lean angle equals speed squared over gravity times the turn radius. There’s no mass in it and no height. A dirt bike and a heavy touring bike lean exactly the same through the same curve at the same speed.

The mechanism is steering into the fall. A tilt with no turn means nothing pushes back, so gravity wins. Steer toward the lean and the path bends into a curve, the road pushes sideways, and that push lifts the bike back up. At 10 meters per second, a tiny nudge of the bars catches a two-degree tilt. At walking pace you’d need a curve tighter than the bike can steer, and standing still nothing helps. Every self-driving planner splits the work the same way: a learned model picks the path, and this formula decides whether the curve is physically possible.

10 minutes. Bookmark & watch today. The one formula that explains why a robot bike fell over at the start line.

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

A car taught itself to stay in its lane in about 15 minutes. Nobody told it where the lane was, and it had no map, just one camera, a safety driver and a score.

The car was a tiny Renault Twizy. The company behind it, Wayve, was started in 2017 by two Cambridge PhD students and has since raised a $1.05 billion round from SoftBank, Nvidia and Microsoft. The car earned points for every meter it drove, and each try ended the moment the safety driver grabbed the wheel. Steering at random, it managed about 7 meters between takeovers. After 11 tries, it drove the whole 250-meter road without a single one.

The trick is one line of reinforcement learning: the value of a move is the reward now plus a slightly shrunk value of whatever comes next. A takeover at the end of a run lowers the value of the moment before it, then the moment before that, until the blame reaches the turn that caused it. In September 2026, Uber and Wayve launched robotaxis in London.

10 minutes. Bookmark & watch today. We run the formula by hand, and the 7 meters vs 250 meters gap makes complete sense.

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

Dean Kamen predicted his scooter would be the fastest company ever to reach a billion dollars in sales. Twenty years later the Segway was discontinued after selling about 140,000 units. The physics inside it never failed.

A rider on a Segway is an inverted pendulum, and the tilt grows on its own. It doubles roughly every two tenths of a second, and in about two thirds of a second you’re on the ground. The machine survives by measuring the tilt and driving its wheels under the fall, again and again, faster than the fall can grow. Walking humanoid robots balance on the same math today.

5 minutes. Bookmark & watch today. Try balancing a broom, then a pencil, and you’ll feel the formula

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

In 1979, a robot needed five hours to cross one room full of chairs. Computers could already play checkers. Seeing a chair was the hard part.

The robot was the Stanford Cart. It rolled one meter, then sat still for ten to fifteen minutes while a mainframe linked by radio, at about a million calculations per second, worked out where everything was. Its builder Hans Moravec later wrote that making computers play checkers at an adult level is comparatively easy, and giving them the seeing and moving skills of a one-year-old is hard or even impossible. That idea now carries his name: Moravec’s paradox.

The cart judged depth with one line of math, Z = f·B/d, using a single camera that slid along a rail and took nine pictures before every move. The same line has a catch. Error grows with the square of the distance, so five times farther means 25 times more error, and at 40 meters a one-pixel slip costs more than 4 meters. Tesla pulled radar from its cars in 2021, then the parking sensors in 2022, and bet everything on cameras. Its cars face the same physics.

10 minutes. Bookmark & watch today. Try the thumb test in the first few minutes and you’ll feel the formula with your own eyes

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

@0xEvrenn ibm left the general purpose facial recognition business in 2020, two years after this audit

[larwick @larwick]

In 2011, a Stanford robotics PhD student and a precision-farming veteran met in a startup class and set out to build a machine that looks at every plant and decides which ones to kill. Six years later, John Deere paid $305 million for it.

Every decision that machine makes is graded by two numbers: precision, how many sprayed plants were really weeds, and recall, how many weeds it actually caught. Turn one knob and you trade one for the other. Today the cameras ride on 120-foot booms, and Deere says farmers used it on more than 5 million acres in a single season.

11 minutes. Bookmark & watch today. The same two numbers grade every AI that has to say yes or no

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

@0xorenx the rods had graphite tips, so pressing az-5 first added reactivity at the bottom of the core before any absorber got there

[larwick @larwick]

@0xzivo every point on the rate moves that 20 year deal about 6%, roughly $7M

[larwick @larwick]

@0xBenjamine skilling reportedly brought champagne to the office the day the sec signed off on mark to market

[larwick @larwick]

in 2015 Oregon State professor Jonathan Hurst founded Agility Robotics to build robots that walk like animals

its humanoid Digit lives by one school formula, runtime = battery energy ÷ average power

the newest Digit works for only about 90 minutes before it walks off to charge

Amazon tested it in warehouses and its maker is going public at a $2.5 billion valuation

a working robot draws about 460 watts, nearly 5 times what your whole body burns

and the twist is a shorter battery can mean more work per day if it charges fast

Agility says it already has more than $300 million in orders riding on that math

11 minutes on the formula plus how to read any robot spec sheet yourself ↓

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

GM Burned $10 Billion on Cruise Before Its $30 Billion Robotaxi Bet Collapsed Without Proving Driverless Cars Were Safe

General Motors put more than $10 billion into Cruise, a robotaxi startup once valued at $30 billion, and in December 2024 Mary Barra shut down GM's robotaxi plans

Cruise had driven about 5 million driverless miles when, in october 2023, a pedestrian hit by another car was thrown in front of a Cruise robotaxi, which then dragged her about 20 feet

California pulled its driverless permit, Cruise recalled 950 cars, CEO Kyle Vogt resigned, and GM eventually stopped funding robotaxi development

the math behind proving safety is the rule of three, zero failures in n independent trials puts the 95% upper bound on the failure rate at about 3 divided by n

so an AI agent that runs 300 times without failing could still fail 1% of the time, and 5 million crash-free miles still leaves room for up to 60 deaths per 100 million miles

to push that bound down to one death per 100 million miles you'd need around 300 million failure-free miles, and real roads make it harder because no two miles are the same and the deadliest edge cases are rare

Waymo, Tesla, NVIDIA and every AI agent from OpenAI, Google or Anthropic face the same wall, a clean demo or benchmark can hide a failure rate you only find in production, and the bigger the cost of failure the more evidence you need

watch this five-minute breakdown before trusting any AI that claims it never fails ↓

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

Disney Built a Spider-Man Robot That Flips 65 Feet Over Crowds Using One Physics Equation That Could Transform Robotics

disney built a robot that launches spider-man more than 65 feet over a live crowd at avengers campus and flips him in midair with no human steering each move

it tracks its own motion with a gyroscope, an accelerometer and a laser rangefinder, and it has only a few seconds to finish the spin, so a tiny error in launch speed changes everything

the whole trick runs on angular momentum, L = Iω, tuck the limbs in and the spin speeds up, stretch them out and it slows down so the robot hits the right pose before it drops out of sight

it's the same move an olympic diver uses, except here it has to work every single show with a machine flying over thousands of people

this stunt follows a carefully engineered flight path, and disney is also training walking droids with reinforcement learning in simulation, the same approach boston dynamics, tesla and nvidia use to teach robots to move

put those together and a robot could practice thousands of simulated launches and learn to correct its own mistakes, which is exactly what robots need to recover from falls in factories, warehouses or disaster zones

watch this breakdown to see how one equation controls a flying robot's spin and why the physics behind disney's spider-man stunt could help shape the next generation of real-world robotics ↓

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

Intuitive Surgical Made $10.1 Billion Turning Shaky Human Hands Into Precision Robots and Johns Hopkins Just Taught AI to Operate Alone

your hand is never perfectly still, even the most experienced surgeon has tiny involuntary movements, but a robot can filter out the shake and turn deliberate hand motions into incredibly precise movements inside the human body

intuitive surgical turned that engineering trick into a $10.1 billion business in 2025, as its da vinci systems were used in approximately 3.15 million procedures in a single year, up 18% from 2024

that is more than 8,600 procedures every day, powered by machines that translate a surgeon's hand movements into controlled movements of miniature instruments

the first trick is motion scaling, if a surgeon moves their hand 3 millimeters, the robot can move its instrument just 1 millimeter, reducing the effect of unwanted movements while preserving deliberate commands

the second trick is tremor filtering, a control algorithm suppresses rapid, involuntary movements while allowing slower, intentional movements through, turning a shaky human input into a much steadier instrument response

combine 3-to-1 motion scaling with effective tremor filtering and a 100-micron hand movement can be reduced to roughly 33 microns before filtering removes additional unwanted motion

this is not magic or a futuristic AI breakthrough, it is control theory, signal processing, and robotics working together to make a human surgeon more precise

but the next breakthrough could be much bigger than stabilizing a surgeon's hand

in july 2025, researchers at johns hopkins university published a study in science robotics describing an AI-powered surgical robot that autonomously completed a complex phase of gallbladder removal across eight experimental gallbladders

the system was trained on surgical videos and used the same broad transformer architecture behind modern AI models like chatgpt, learning to identify anatomy, manipulate instruments, and recover when its actions did not go exactly as planned

the researchers reported a 100% success rate across those eight ex vivo experiments, a major research milestone, but not proof that an AI robot is ready to perform entire operations independently on living patients

now look at the business opportunity, intuitive surgical generated $10.1 billion in annual revenue while researchers are pushing surgical robots from tools controlled by doctors toward systems capable of executing increasingly complex tasks on their own

the companies that master this transition could reshape a global medical industry worth trillions, although regulatory approval, patient safety, and real-world clinical evidence will determine how quickly autonomous surgery reaches hospitals

the same two ideas that make a robot hand steadier today, precise motion control and intelligent filtering, are becoming part of a much larger race to build machines that can perform delicate physical tasks with human-level adaptability

watch this to understand how a robot cancels a surgeon's hand tremor and why the next $10 billion robotics business could be built around AI that learns to operate, not just assist ↓

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61

[larwick @larwick]

NASA Spent $85 Million to Fly a 4 Pound Helicopter on Mars and It Crushed Its Mission With 72 Flights

NASA spent roughly $85 million building and flying Ingenuity, a 4 pound helicopter that was supposed to fly five times on Mars, and then watched it fly 72 times in air about 100 times thinner than Earth's

nobody had ever flown a powered aircraft on another planet, and the Martian air is so thin that a normal helicopter would barely get off the ground

one equation explains why, lift = ½ρv²SCₗ, where air density, rotor speed, blade area and lift coefficient decide whether anything flies

drop the air density to 1% of Earth's and you lose about 99% of your lift with everything else the same, so NASA had to push every part of the rotor harder

the blades spun at around 2,400 rpm, way faster than a normal helicopter, and the whole thing flew itself because Earth was sometimes 140 million miles away

the onboard brain was a Qualcomm Snapdragon, the same family of chips you'd find in a smartphone

thin air was only half the problem, Mars gets brutally cold, solar charging is limited, and every flight had to fit inside a tight energy and temperature budget

NASA planned five flights as a tech demo and Ingenuity made 72 before rotor damage grounded it for good in January 2024

now NASA's Jet Propulsion Laboratory is using Anthropic's Claude to help plan real Perseverance rover drives

for SpaceX, NVIDIA and everyone building autonomous robots the takeaway is the same, a smart machine still has to make its decisions inside hard limits on physics, energy and hardware

watch the physics behind the $85 million helicopter that flew 72 times on a planet where flying shouldn't work

[larwick @larwick]

wall street is throwing $5 trillion at robots that still can't fold a basket of laundry

investors price in a sci fi future for machines that struggle with your fucking chores

read this before you become exit liquidity for the biggest robot hype trade ↓ https://t.co/Pw34Jrnr61