NYU's eFlesh is a $5 tactile sensor you 3D print yourself
Four parts, a hobbyist printer and under five dollars of magnets, with all design files and trained models released openly.
Lukas Ziegler flagged eFlesh, an open-source magnetic tactile sensor out of New York University that costs almost nothing to build. The parts list is four items. A hobbyist 3D printer, off-the-shelf magnets for less than $5, a CAD model, and a magnetometer circuit board.
The idea is simple enough to explain in a sentence. The sensor is 3D printed with magnets embedded in a middle layer. When something presses on it, the magnets shift, and a magnetometer underneath reads that displacement as contact force. How much they shift depends on the mechanical properties you choose when you design the part.
Print the shape you need
The part that makes it more than a one-off lab gadget is the design tool. It converts plain OBJ or STL files into 3D-printable STLs, so you can make a sensor shaped for the thing you actually have. Ziegler lists robot hands, grippers and quadruped feet as examples. Most tactile sensing work ends up locked to one specific finger or pad geometry, which is a big reason the field has stayed fragmented and a lot of manipulation work is done sensorless.
The numbers
The writeup reports slip detection that generalizes to unseen objects at 95% accuracy. Visual-tactile control policies, meaning policies that get both camera images and touch, improve manipulation by 40% over vision-only baselines. On precise tasks like plug insertion and credit card swiping, success hits 90%.
Those are exactly the tasks where vision alone tends to fall apart, because the last millimeter of a plug going into a socket is usually hidden by the gripper itself.
Shown versus claimed
All of these numbers come from the project's own writeup, not an outside benchmark or a third-party test. Nobody outside the group has reproduced them yet. The useful part is that the barrier to checking is unusually low. Ziegler says all design files, code, trained models and conversion tools are openly available, and the hardware bill of materials is a printer plus a handful of magnets. If the slip detection number does not hold up, somebody with a Prusa and a free weekend can say so.
That combination is rarer than it sounds. Plenty of tactile sensing papers publish strong results on hardware you cannot buy, which means the claim sits unchallenged for years. Here the cost of disagreeing is five dollars.
Slip detection generalizes to unseen objects with 95% accuracy. Visual-tactile control policies improve manipulation by 40% over vision-only baselines
the lack of a versatile, accessible tactile sensor has led to fragmented solutions and often force-unaware, sensorless approaches