Building a tendon-driven robotic hand with GPT-6 Astra
The author calls it his hardest Astra project yet: 7 actuators, 16 articulated axes, tendon-driven fingers and an articulated thumb, all inspectable in an exploded browser view to check for interference.
This might be the hardest thing I’ve tried to build with GPT 6 ASTRA so far. A robotic hand looks simple until you start trying to make every tendon, pulley and actuator work together. The current version has 7 actuators, 16 articulated axes, tendon-driven fingers, return springs, cable routing and an articulated thumb. And every revision exposes something new. Move a pulley a few millimeters and the tendon path changes. Give one actuator more clearance and suddenly the thumb has less room. Fix that and another cable starts rubbing against the geometry. I can now inspect the entire mechanism in an exploded browser view and move the joints to see where things start going wrong. There’s still a lot to improve, especially the kinematics and grasping. Next I want to add tactile sensing and push the grasp simulation much further. I was also in the kitchen. Robotics HumanoidRobots PhysicalAI DexterousHands
This post quotes another post: x.com/techniahqrobot/status/2098491528045821964