Sharpa Robotics just dropped a new hand video and the level keeps going up. This is the Sharpa Wave running WM Craftnet on a human scale fivefinger hand with 22 active DoF. The policy combines wrist depth, tactile sensing, proprioception and previous actions. The hand can rotate different objects... in-hand, recover after external pushes and continue manipulating objects it was never trained on. The numbers are strong. 175/200 successful real-world rotation trials across 20 objects. A world-model prior trained on 9 objects was transferred to 49 new objects. Fall rate went from 6% to 0.3%. The Wave hardware itself has 22 actuators, up to 20 N fingertip force, 240×240 tactile sensing at up to 180 fps and 0.02 N pressure sensitivity. What caught my attention is the recovery behavior. The fingers keep changing contact points after the object slips or gets pushed instead of replaying the same finger motion. That is the kind of dexterity I want to see more of in robotic hands. According to Sharpa’s current specifications: • DTA tactile sensors on the fingers with a resolution of up to 240 × 240 • Pressure detection • Slip detection • Force change detection • Contact point localization • 6-axis force and torque measurement: Fx, Fy, Fz, Mx, My, Mz • Tactile sensing at up to 180 fps • 20 ms reported latency • Force detection range from 0 to 30 N • Maximum sensor load of 50 N • Sharpa also describes a miniature camera integrated into each fingertip for visuo-tactile sensing.show more

So these are sensors? At some point I imagine robots will have finger prints for identification and grip.

Yes, those are tactile sensors. I could definitely see robotic fingertips getting textured surfaces similar to fingerprints too. The ridges could help with friction and give the sensors richer contact information during grip and slip

Humanoid robots are getting very good at walking. But I’m starting to think hands will decide who actually wins this race. If the robot can’t reliably manipulate everyday objects, what’s the point of perfect locomotion? Hands or mobility — which one is the real bottleneck?

I’d put more weight on hands right now. Walking is already reaching a usable level on several humanoids. Reliable grasping is still much harder once you add different shapes, soft objects, cables, tools, slip, force control and contact changes. A robot can walk across the room perfectly and still fail on the door handle.

Exactly. The door handle example says it all. 😂 Walking gets you to the task. Hands decide whether you c an actually do it. What do you think is the hardest everyday object for a humanoid right now?

The WM-Craftnet paper clocks its tactile stream, wrist camera and hand controller at 10 Hz. Sharpa rates the fingertip sensor at up to 180 fps. The sensing is quicker than the loop it feeds, and I'd like to know what sets the pace.

The 180 fps figure is the sensor’s maximum capture rate while the policy runs at 10 Hz. My guess is that tactile samples are being downsampled or aggregated before each control step. I’d like to see whether inference, camera rate, communication or sensor fusion is setting that 10 Hz rate.

The rate table puts 10 Hz under hardware rates: hand control, the wrist camera and the tactile stream. The sim steps physics at 60 Hz and still keeps that control rate. Joint targets get smoothed by a filter before execution, and the depth camera there is a RealSense L515.

Reports put the fall rate about 6% to 0.3%. Fingers keep hunting new contact after a shove instead of replaying the same move. 22 DoF. 240x240 tactile at 180 fps. World model prior on 9 objects, then 49 it never met. Hands were the scarce part. This is how they stop being rare. @techniahqrobot @SharpaRobotics @Figure_robot $PATH $AMD #Robotics #PhysicalAI #Abundance

A hand that can detect slip and search for a new contact after a disturbance is much closer to useful manipulation than one that only repeats a learned trajectory.

Impressive

Mind blown! The Sharpa Wave's capabilities are insane - 175 out of 200 successful trials is nuts!.