Simon Yu

@suren_at What was the main chip? We just made a teensy based breakout board

weekend update: a new custom component for the robot Try to guess what it is 👀

@dolylupec @wuji_global @TheHumanoidHub @XRoboHub I thought worm gears aren’t back-drivable. We have the first version of the Wuji hand, and those can’t be back-driven. I’m curious what they modified to make it back-drivable

Super excited to visit @wuji_global who's pushing the frontier with their advanced dexterous robotic hand Most traditional robotic hands still rely on tendons/cables, but Wuji Hand uses direct-drive actuators embedded right in the fingers with worm gears. This solves the usual tendon problems by delivering smoother, more precise, and reliable motion with way less sim-to-real gap It also has high dexterity with 20 active degrees of freedom and the weight/size feel almost exactly like an adult human hand Got to test the newly launched Wuji Hand 2 today and was super fascinated! (watch how I handle the hand in the first video😂)

@liminalsunset_ @boxcardavid Recently it’s about firmware, sat up the can id through their software, and after a few on and off power cycle while doing experiments on our robot, some actuators got disconnected, later they found that the can ID get reset to 0 and all things appears to went back to default set

@SimonYu_Chinese @boxcardavid what part of them specifically is unpredictable for you? is it the software or the hardware reliability?

They’re using Encos actuators, which are excellent. Damiao and Encos are usually the right answer (for now).

The first version of our custom motor controller board is here. Let’s see how far we can take it! 🚀

@cixliv 😅

Using a mocap suit to kick yourself in the balls with a robot is a great metaphor to close out 2025.

1st version will probably stick to can, later would like to try ethercat, anyone familiar with the communication protocol?

@CHENYAO981123 Hi Chen, we are considering to buy some my actuators, may I have you WeChat? Thanks 🙏🏻

The X12 sounds splendid, but the noise generated by planetary gearboxes remains a persistent challenge for actuator manufacturers.

@Sentdex RL with reference trajectory, probably similar to beyond mimic paper

This might be the best humanoid jog I've seen yet. Doubt this is pure RL, I'd love to know the curriculum here.

@observie Hi David, is it a Pic chip? Also were you able to find Xt30 2+2 in the US, it looks like it’s only available through Chinese vender like jlcpc.

- Using two Würth Red Cubes now instead of the AS150U - Fixed footprint of current sense resistor WSL3921 0.1 mΩ to match the datasheet - Switched to IPT010N08NM5 for extra 80 V margin, accepting slightly higher resistance of 1.05 mΩ (vs. IPT007N06N's 0.75 mΩ), but still a big win over the original P-FET

@Manh_Ti3012 I am curious how you trigger your adc reading to ensure low side fet is on, pwn interrupt? do you use pin that can do triple simultaneous mode? Is the reading for the 3 add up to zero? Also beautiful board :)

@SimonYu_Chinese The driver have 3 current sense amp, so i just use all 3, save some headache with current calculation

Not a fan of dual side component but have to do this for the optimal layout😓 BLDC FOC Driver board

@cixliv On one end, we know safety is crucial; on the other end, making a robot completely safe before deploying it may give away the market to competitors on the other side of the world who are already deploying large numbers of robots to their customers.

Well the good news is that Figure would be a good fighter. Not all bad news. Hit me up Brett!

@GoingBallistic5 @observie @KWRoboticsAI I talked to the people who went to IROS, and he said they use thick steel cable for the ankle roll.

@observie @KWRoboticsAI It's a mystery. The ankle is not tradition RSU. It is quasi-serial, no longer parallel actuation. The roll could -- stress on could -- be compliant with two control rods either attached to opposite sides of an actuator, or damping springs. I don't think tendons though.

Even the foot of a humanoid robot is not straightforward to design. Tesla's Optimus has active ankle pitch and roll DOFs, controlled by two linear actuators, and an articulated (passive) toe box (coined by the great @GoingBallistic5). Unitree's H2 has an active ankle pitch, and an ankle roll joint which seems passive. No visible toe box. Xpeng's Iron has an active pitch and roll ankle joint, which seems similar to Optimus's. No visible toe box. The K-bot has active ankle pitch, no roll, no toe box. The K-bot approach is the simplest and least capable, which is the reason I like it so much. I hope to get a humanoid up and running with lots of mistakes to learn from - and then improve and iterate. But I can't stop thinking about the options.

@cixliv 😅

@SimonYu_Chinese Lex?

@gan_chuang Thought they will go all out on robotics, didn’t they just recently hire Prof Sangbae Kim 😲

Meta has given up on driving foundational innovations in robotic learning...

@ORHEX0 Found Berkeley humanoid lite use the exact ESC, you can try to use their code github.com/T-K-233/Recoil…

@SimonYu_Chinese Not yet just got it I’ll write the software on my free time and uploaded to my GitHub

YEAH BWOI

@KyleMorgenstein Whether you like it or not, your foe from authoritarian regimes will always find someone to do it, and they will spare no recourse.

If you aspire to weaponize humanoids, then our visions of the future are fundamentally incompatible.