Fog you can sit on

In 1993 a computer scientist named J. Storrs Hall told a NASA conference to fill a room with robots. Small ones. Each would be about the size of a human cell, a hundred microns across, with twelve arms, and a bucketful of them would link arms into a crystal. Fill the air of a room with them and the room looks as if fog has rolled in. Hall called it Utility Fog. The paper is twelve pages on NASA’s server, and it thanks Carl Feynman, Richard’s son, for checking the numbers. This is the first note under nanotechnology because it is the clearest statement I know of what the field is for, and because thirty-three years is long enough to ask how much of it arrived.

What the fog does

The trick is in the programming, not the robots. In the plain mode each robot is a cell of some object: it holds a position and does a job. In fog mode the robots are pixels. The object is a pattern that moves across the cloud while the robots stay where they are, the way a picture crosses a screen whose pixels never move. A chair is a region of fog told to push up on the seat of your trousers with the pressure a chair would. Walk away and the region dissolves.

Most of the room is simulating air. The robots that stand for empty space run a fluid simulation of the air they displace and do what that air would do, so you can walk through them, and breathe. They take up a tenth of the volume. The rest is air. Hall draws the limits as carefully as the powers. The fog cannot be hot, cannot be as hard as cold steel, cannot do chemistry, cannot be eaten, and to the eye it looks like heavy fog, so you wear goggles. It is a universal effector and a universal sensor for everything between those limits: shelter, clothing, telephone, furniture, car, all appearing when wanted. The space section is the one he wrote the paper for. Fill a cabin with fog and nobody floats out of reach. It is an acceleration couch when the engine is on and empty space when it is off. Put it inside a spacesuit to manage the pressure, outside the suit to give fine hands, around a station as shielding and as a tug.

The five numbers it needs

Hall writes down what he assumes, which is why the paper is still worth reading. Five numbers carry the whole thing.

What a foglet is, by the paper

The body. A hundred microns across, twelve telescoping arms on the faces of a dodecahedron, a three-fingered gripper on each. Arms link into Buckminster Fuller’s octet truss, a face-centred cubic lattice where every robot touches twelve neighbours. The skeleton is aluminium oxide, not diamond, because a diamond fog with that much surface in the air would be a fuel-air explosive. The bulk has the density of balsa, a fifth of water, and the tensile strength of polyethylene. Retract every arm and it contracts to the density of water.
The brain. A thousand MIPS per cubic micron, a 1993 supercomputer in every foglet, built from Drexler’s mechanical rod logic. Sixteen million foglets to the cubic inch, a trillion to the cubic metre.
The muscle. Electric motors at a milliwatt per cubic micron, arms moving a hundred thousand times a second at ten metres a second. Pushing arms spend energy and pushed arms recover it, so the budget nearly balances, with hydrogen in each foglet as the buffer.
The nerves. Optical links between neighbours. Passing messages hop by hop would crawl at fifty metres a second, so the fog needs switched circuits through itself.
The factory. Foglets do not reproduce. A machine the size of a breadbox makes them, by molecular manufacturing, Drexler’s atom-by-atom assembly.

Set those five and the rest is arithmetic, which Hall does. A column of fog lifting a person carries a thousandth of its rated load. Sliding layers past each other at full load would burn 700 kilowatts per cubic centimetre, the power of a pistol cartridge, which is why the motors must be generators too. The computers alone, at a few nanowatts each, make kilowatts of heat per cubic metre, so the fog must cool itself by pumping its own air.

A room full of fog. A small room, four metres by four by three. Set the spacing between robots and the price of one, and read how many it takes and what they cost. The presets are the only prices that exist: a micromirror from a cinema projector, which is a ten-micron machine made by the million on a wafer, and a centimetre module from a robotics lab. The mass line does not move because Hall's fog has a fixed density whatever the robot size.

Where each number stands in 2026

Hall assumed the machines would be built from the bottom, atom by atom. That has not happened. Nobody has a mechanosynthesis assembler, and the field that was going to deliver one spent the 2000s arguing about whether it could exist. What arrived instead was molecular machines in chemistry, a Nobel Prize in 2016 for motors and shuttles that make one motion in solution, and DNA origami that folds into shapes but does not push. Neither is a foglet.

The surprise is that the foglet came from the top. The lithography that printed chips at 800 nanometres in 1993 prints them at 3 nanometres now, and the same tools make machines.

The body is done, the arms are not. In 2020 a Cornell group printed robots a hundred microns long, a million to a four-inch wafer, with legs of platinum seven nanometres thick that bend when a voltage is applied, powered by light falling on photovoltaics on their backs. They walked. In 2022 the same group put a CMOS circuit of about a thousand transistors on a robot a quarter of a millimetre long, and it walked on its own clock. In 2024 they made robots two to five microns across, driven by magnets. The size is solved. The twelve arms are not. Hall’s arms telescope at ten metres a second and hold seventy millinewtons each. Today’s microactuators bend at hertz to kilohertz, move microns, and push with micronewtons. To hold a sitting person each arm needs about seventy micronewtons, which is within reach. To be as stiff as Hall’s fog it needs a thousand times that.
The brain is the easy part now. Hall wanted rod logic because 1993 silicon was too big. Today a square of logic a hundred microns on a side holds about a million transistors, a full microcontroller, roughly a 1990 desktop computer. He asked for a supercomputer per foglet and will not get it. He will get more than enough to run an arm and a message router, which is all the paper actually uses.
The muscle is a hundred million times short. A milliwatt per cubic micron is ten to the fifteen watts per cubic metre. The densest microactuators made, piezoelectric stacks, manage about ten to the seven. Nothing on any roadmap closes that gap, and nothing needs to for the chair. It matters for the stiff fog, the fog that cuts and lifts.
Power delivery is the real wall. A trillion robots per cubic metre have to be fed. Light works for a layer on a wafer and stops at the second layer of a fog. Hall’s answer was power lines through the grippers, so every link in the lattice is also a connector. A trillion connectors mating a hundred thousand times a second without wear is a problem nobody has started.
The nerves are where 2026 is strongest. Mesh networks, swarm algorithms, and data centres of a million machines that stay coherent are ordinary now. Harvard’s Kilobots formed shapes with a thousand robots in 2014. A trillion is a million times a million, and the switched circuits Hall asked for are routine in fibre. The software is the part I would bet on.
The factory exists, at the wrong price. The micromirror chip in a cinema projector holds millions of machines ten microns across, each on its own hinge, each tilting thousands of times a second, for a hundred dollars a chip. That is a sparse fog in two dimensions that does one trick. At that price per unit the small room in the figure costs about a billion dollars and weighs ten tonnes. The price has to fall a thousandfold before a house-sized budget fills a house.

The next twenty years

The city of fog will not be here by 2046. It rests on two things with no roadmap: atomically precise manufacture and motors a hundred million times denser than any made. What can be here is something like it, in three sizes, and I think it comes in this order.

Fog in a tray, this decade. A few million to a hundred million microrobots in a dish, in liquid or in air, with a microcontroller each and light or a floor electrode for power, forming shapes and pushing on things. This is the direction the Cornell and Penn groups are already walking. The first uses are small and worth money: a haptic surface that becomes any texture under a fingertip, a cell-sorting floor, a wound dressing that moves. It is Hall’s fog with the room removed and the goggles not needed.

Coarse fog, the 2030s. Robots at a millimetre to a centimetre, not a hundred microns, which are cheap to make and easy to power, forming furniture, scaffolding and jigs. This is the industrial fog Hall sketched, in factories and in space, where a cubic metre of smart ballast in a cabin earns its mass as a couch, a wall, and shielding by turns. You cannot breathe through it. You can stand on it. Modular robotics has built the centimetre version in labs for twenty years; what changes is a chip and a battery in every cube for the price of a toy.

Fine fog, the 2040s, worn before it is lived in. The hundred-micron robot with arms arrives when a power-carrying lattice exists, and the first fine fog will not fill a room. It will fill a glove, a sleeve, a boot, the inside of a spacesuit, where the volume is litres, the host supplies the power, and the job is pressure and touch. That is the one application in Hall’s paper where the mass, the cost and the power all fit at today’s trajectory. The room comes after the glove.

What Hall got right is the shape of the idea: a computer in every grain, objects as patterns rather than things, and a precise list of what the fog cannot do. What he got wrong is the road. Feynman said there was plenty of room at the bottom and the people who went there went with chemistry. The ones who built the foglet came down from the chip. That is the lesson I take for everything else under this tile. The scale is the goal. The route is whichever one is already paved.