#video by #Harrison-Low on his jugglebot triumph. He finally switched from kevlar kite string to #UHMWPE for the linear actuators. Also he switched to #ODrive Pro electronics to control the motors. Featuring a #JLCPCB #3D-printing ad, sigh. Oh, also he burned out the hand motor because a sensor cable came unplugged. #robotics
on 02026-08-18#Harrison-Low’s Zulip site for his #robotics work #toread
on 02026-08-18#video of #Unitree’s new humanoid robot prototype doing wuxi tricks. #robotics
on 02026-05-17#video of a New Year #robotics show in #China thanks to Sculptor. The robots are child-sized humanoids, performing a martial-arts form including backflips, on stage on a specially prepared smooth floor. They are spaced far apart, and of course the performance is pre-scripted, because martial-arts forms are pre-scripted by definition. They performed with humans who perturbed them somewhat, grabbing staffs they were swinging. Toward the end the robots performed with nunchaku and did a bit of breakdancing and parkour.
on 02026-02-16#paper on a paramecium-sized programmable, autonomous robot that integrates sensors. “We operated within a limited power budget of ~100 nW by building our robot in a 55-nm [CMOS] process and leveraging #subthreshold digital logic. (...) In 210 μm by 340 μm by 50 μm, we were able to fit photovoltaic (PV) cells for power, sensors for temperature, four actuator control circuits, an optical receiver for downlink communication and programming, a processor, and memory.” But only barely: “The processor consumed most of the power, requiring nearly a third of the robot’s area to be devoted to energy harvesting to compensate. Even when using larger, low-leakage transistor-based memory, at this process node, we were limited to storing a few hundred bits.” 23 11-bit instructions, and the program could only be 32 instructions long; it had 16 bytes of RAM, 4 8-bit registers, no memory indexing, and no subroutine call and return functionality. #MEMS #robotics #toread
on 02025-12-18#video #toread about voice coil #robotics #hardware #mechanisms from #Breaking-Taps. Inspired by birds. Using Dyneema cable actuation. Glass-filled PTFE bushings, carbon-fiber bones pressed into holes in fittings. He actually gave up on the rolling-contact cable-ligature joints and switched to conventional sleeve-bearing pin hinges. Using voice-coil actuators.
on 02025-10-14#video by #Boston-Dynamics #robotics about their new three-finger opposable-thumb gripper.
on 02025-10-08#video #toread about folding #robotics actuator mechanisms using an endless cable around two pulleys (with multiple turns of the cable around the smaller pulley) into a very compact package with four idlers.
on 02025-09-03#video on GPS-denied, anti-jam autonomous #drones and #motion-planning using a #Raspberry-Pi 4 running #ROS. He calls his flight control software #dRehmFlight. Apparently Intel’s “RealSense tracking cam” does “optical flow” inside the camera to measure changes in orientation. The main drone body seems to be laser-cut from MDF. He’s using #D-star Lite instead of A* for his #path-planning graph search. #robotics
on 02025-02-20#robotics biped by Qihao Guo. #video
on 02024-10-26#robotics #video by researcher peng-zhihui in China who built a robot arm with various different user interaction modes, running on LiteOS #HarmonyOS.
on 02024-10-26#Harrison-Low #video about how his #Stewart-platform juggling robot finally can juggle, two balls anyway. He actually did switch from kevlar to #UHMWPE (Dyneema) which I thought wouldn’t handle the friction heat. He shows the pseudo-capstan “hand” #mechanisms that actually throw the juggling ball. Oh, and he's using Zulip instead of Discord to make it more open. #robotics
on 02024-09-15#Harrison-Low #video #toread about redesigning #robotics joint #mechanisms
on 02024-09-15#Harrison-Low #video about eliminating knotted string from his #robotics #mechanisms by wrapping his kevlar threads around screws and then squishing it under the screw heads by tightening them. Maybe he should have just used a trucker’s hitch?
on 02024-09-15#Coy-Beardmore #video on salvaging industrial robot arms. Holy shit, US$80 for a cleanroom #SCARA arm. And then a guy who has a warehouse full of used robot arms gave him another one, plus the power supply and control electronics, for free. So then he spent a bunch of time with a vertical milling machine making solid steel rails to mount the robot arms on. Adept V+ is running in a windowing environment I don’t recognize, on some kind of OS with drive letters. Software dates 01984-01995, supporting TCP/IP, and V+ is apparently from 02000. He had to troubleshoot a short circuit in the wire to the Z-axis home sensor, whose insulation had frayed; it kept blowing a 2-amp fuse when the switch closed. There’s some kind of weird bespoke programming language visible at 4'56". #hoarding #robotics
on 02024-06-08#video about improved high-speed linear actuator #mechanisms design using carbon fiber for #Harrison-Low's juggling robot. Now his motor is in the linear actuator itself, and its shaft has two reels on it, one to let out string (kevlar kite string?) and one to take it up, or vice versa when moving the other direction. He’s also added TPU sleeves to the six bearings he was using as rollers to constrain the shaft’s motion to one degree of freedom. I think it's about 500mm of stroke. He’s able to get 3.4m/s and >50k cycles endurance. The theoretical precision is 8.4μm, which is more precise than his ±20μm Mitutoyo calipers, but with a pulley mechanism he got a standard deviation of 57μm, which is pretty good. The string breaks at about 70 or 80 newtons. #robotics
on 02024-04-28#Harrison-Low #video about a #mechanical linear actuator design for a juggling robot, capable of a full 263mm stroke in 100ms. He’s mostly using carbon-fiber reinforced plastic tubes, with six 6001RS cartridge bearings for rollers (one too many for #exact-constraint), held in place by a piece made by #3D-printing, with interchangeable spacers to adjust the fit in 100μm units to the imprecisely-sized tubes. He’s doing the actual actuation with a kevlar-kite-string cable drive, with the cables actually run through teflon Bowden tubes so the motors don’t have to add mass to the actuator joints; the motor pulley has two V grooves for the two ends of the cable. This seems like a good way to do robotic #mechanisms in general. This replaces his first hydraulic design made out of syringes and a second iteration using a ballscrew which was too slow and (I assume) also super expensive. He has an AS5048 magnetic encoder for positional measurement, coupled to the actual moving parts with a constant-force spring, whose rotation is what it measures, but so far he’s just using encoders on the motors themselves (which I guess are brushless since he’s using #ODrive; in fact, the motors are ODrive-branded, model D6374-150KV). He says ODrive's current sensing “completely removes” the need for limit switches. His attempts to clamp the three frame tubes with a flexure from the inside failed, but I don’t know why he doesn’t clamp them with a flexure on the outside, instead relying on hammering them in. He uses a thin wire with a hook bent on the end as fish tape to pull the kevlar string through the teflon Bowden tubes, but doesn’t always need it because of how stiff the kevlar string is. On one occasion the kevlar string was able to friction on the 3-D printed plastic and just cut right through it, presumably from heat (a big advantage for kevlar over UHMWPE, except when you want to cut it; presumably kevlar will creep less for the same reason). #robotics
on 02024-04-28A #paper on “imitation learning” in #robotics with #deep-learning
on 02017-08-06#deep-learning #neural-networks with #tensorflow for low-cost #robotics
on 02016-09-23#paper #pdf #introduction to the SLAM (“simultaneous localization and mapping”) #robotics problem and the #algorithms for solving it, focusing on extended #Kalman-filters using a range measurement device. SLAM is the process of building some sort of model of the world your robot is in while also figuring out where you the robot is in that world. Unfortunately, this overview is from 2005, more than ten years old now, before the particle-filter revolution.
on 02016-01-11on #Kalman-filters in #robotics. Doesn’t even mention #particle-filters!
on 02015-08-12