@physicalai.robot: An open-source robotic hand that can sort an object it has never seen before using touch. Researchers at Carnegie Mellon University and FAIR at Meta AI developed D’Manus, an open-source robotic hand designed to keep sensing an object even when vision becomes limited. It has three fingers and 10 degrees of freedom, driven by Dynamixel XM430 actuators. The contact surfaces use ReSkin, a soft magnetic material that allows the hand to detect contact. There are 56 magnetometers beneath the skin. Eight are placed in each fingertip, with another 32 across the palm. When an object presses against the surface, the sensors detect changes in the magnetic field. A trained model uses those signals to recognize properties such as softness and texture. The video illustrates the hand sorting objects through touch, including an object it had never encountered during training. It also shows how contact sensing can remain useful after the fingers close and the object becomes hidden from view. The researchers report 10,000 operating hours across versions of the platform, with five breakages. The open-source design includes 3D-printable components, STL meshes, CAD files, assembly instructions and a MuJoCo simulation. The estimated cost of the original parts is around $3,500. I like that the researchers placed 32 magnetometers across the palm. The hand collects tactile information even after its fingers close around an object. #Robotics #OpenSource #OpenSourceRobotics #PhysicalAI #RoboticHands

physicalAI humanoid

Open In TikTok:

Region: MA

Saturday 14 November 2026 08:40:42 GMT