Introducing Muscle v0 -- infinite degrees of freedom, from @DaxoRobotics. A different mountain to climb - with a far more beautiful peak. We built this from the ground up: - Ultra-dexterous - Built for machine learning - Durable and robust More below (1/n)

This could shock some markets including humanoids. What an exciting time to be alive!

A consumer robot led by two of the most influential authors in physical ai. Plus full vertical integration and a team full of stars. Will robotics be solved? Nov 19th everyone 🔥

The amount of work the build team has done is astounding. Congrats on the launch.

A loss of propriocetion seems to have bigger impact on general locomotion & manipulation skills than a loss of touch/pain. Touch/pain sensing seems to be crucial for the safety of our fingers. Both form of limitations can be compensated with vision.

Watching how people with different limitations manipulate objects feels like an accidental ablation study. It’s incredible how our senses compensate for each other. Some of these examples may challenge what we thought was “essential” for manipulation. Man who lost propriocetion:

A scrappy recap on muscle v0. We have gotten many great questions and requests in the past two months. Now we are moving into an intense build mode. To learn more about our unique approach, check out the podcast from @RoboPapers hosted by the awesome @chris_j_paxton @micoolcho.

This is a portion of the force info we get from the hand, both when it's actively controlled and passively wiggled. We damaged some motors (black hexagons) from shipping the robot. But the fingers still have their full range of motion, thanks to the robustness of complexity.

Ilir’s podcast is different. He’s as curious about the founder as he is about the tech. I always wanted to build something, but it took many years to learn how. Grateful for everyone who helped along the way. If you’re thinking of starting something but stuck, hope this gives you

It turns out, being able to bend my DIP independently is a sign of better neural independence, and perhaps slight variation in my tendon connections. Would the dexterity of the hands of hand builders influence the hands they build? @GoingBallistic5 @anatomyumea

1/ The future of general-purpose robotics will be decided by one major question: which flavor of data scales reasoning? Every major lab represents a different bet. Over the past 3 months, @adam_patni, @vriishin, and I read the core research papers, spoke with staff at the major

Years ago, the book "Dynamism: The Values that Drive Innovation, Job Satisfaction, and Economic Growth" broadened my perspectives on what it really means to improve productivity through technology. The last two chapters specifically talked about the effect on wages by robots that

Roboticists should read social science. The mass adoption of robots to unlock productivity is both a technical and social problem. Nobel Prize winners in economics talk about "creative destruction" and the mechanism behind tech advances. Will deep dive into their work soon.

Zero load test is a standard benchmark for robotic hands. The "number of cycles" specified by robotic hand datasheet generally means zero load test. To do a loaded test, it really depends on the use cases, which can have different loading conditions.

100k cycles of finger open & close. No motor overheating. Very little wear on the finger tendons. Tendon-driven hands don't usually publish their data on fatigue because tendon wear & tear is an unsolved problem. But we found a way.

Could be useful for dexterous manipulation using robotic hands if occlusion can be solved and meshes extracted?

Notice how Apollo's wrist camera is underneath the hand. Our tactile sensing likely helps to cope with vision being blocked during manipulation. A wrist camera that sees hand–object contact can reduce this need, which explains why simple parallel jaw grippers work well using

Just another day at @DaxoRobotics. One thing we observed is that the precision of the finger improves as we add more motors to it. Couldn't find any theory behind it (because nobody builds robots like we do) We initially thought it's just that noises could cancel each other out.

A 12mm diameter coreless motor with planetory gear costs ~$15 at 1000 pieces. A 10mm one costs ~$65. That's an insane 333% jump in cost for just 2mm. To build a robotic hand, the marginal cost for miniaturization has outpaced that of complexity. We need a new path forward.