a humanoid robot starts with a ten minute video
i watched a berkeley team assemble a humanoid robot actuator out of printed plastic, in a video under ten minutes long. i want to build this at home, and i'm looking for people to build it with.

today i watched a nine minute and forty five second video four times in a row.
it's a tutorial from the berkeley humanoid lite team. no pitch, no keynote, barely any talking. a pair of hands assembles one actuator, the joint that moves a robot's limb, on a workbench.
you clean the support material off the printed parts. you press threaded inserts into the plastic with heat. you push bearings onto a shaft. you mount two cycloidal discs, offset by 180 degrees, and check that they spin freely. housing, motor, encoder on the back. the narrator says yay, and that's the whole video.
watch it here. it will change what you think a humanoid robot is made of.
i spent eight years at quanser building robotics and control systems for university labs. back then a gearbox was a precision part you ordered from a catalogue and waited for. in this video the gearbox comes off a desktop 3d printer.
the robot these actuators go into is about 0.8 metres tall and weighs 16 kilograms. it has 22 actuators. the team's own bill of materials comes to 4,312 us dollars if you buy the parts in the united states. the hardware designs and the code are open, so you can read every file tonight.
here's the part i didn't expect when i went through that bill of materials. by my count the printed plastic is about five percent of the cost. the actuators are about eighty percent, and most of that is motors, bearings and electronics you buy. so printing is the cheap part and the easy part. the real work is in that video.... somebody sitting down and pressing bearings, lining up keys, bolting on an encoder, more than twenty times over.
which gave me an idea i can't put down.
what if the goal was an open humanoid that can help build the next one?
not all at once. first a printed hand that can lift a finished part off the printer bed and start the next job. then a robot that sorts the parts for one actuator. then one that presses the bearings in. then a whole actuator, then a limb. i have looked, and i have not found anyone showing a humanoid assembling its own actuator. plenty of them move boxes. my search wasn't exhaustive, so if you've seen it done, send it to me.
so i'm going to build this. a 3d printer is on its way to my house. i'm starting small, with one finger of an open source hand, then the hand, then an arm, and working up to a berkeley humanoid lite class walker. i'll post what works and what breaks.
i'd rather not do it alone, and i can't do all of it well. i'm looking for a few people. someone who has printed and run cycloidal gearboxes and knows how they wear. someone who lives in motor control and embedded firmware. someone who trains walking policies in simulation and has got one onto real hardware. someone who has done teleoperation. toronto area is a bonus because parts are easier to pass around in person, but i care more about what you've built than where you live.
this is a personal build. there's no company behind it and no salary to offer. if that's you anyway, email me with one thing you've made that moves.
and if it isn't you.... what would you build first if the gearbox came off your own printer?
the machine economy brief
one email when it matters: bitcoin, ai, robotics, and what founders should do about it. unsubscribe anytime.