MIT researchers build sensing into 3D-printed objects, from robot gloves to lamps

In an October 22 account, MIT detailed how researchers at its CSAIL laboratory and Tianjin University used X-Hinges to build these touch-responsive objects.

The new account describes demonstrations and a calibration workaround for research first submitted to arXiv on September 10. The paper is accepted at ACM UIST 2026, and the team plans to present it in early November. This is a design and fabrication research project, rather than a newly announced commercial rollout.

Separating the sensor from its signal path

The problem starts inside the printed material. Earlier approaches often used the same conductive material for both sensors and traces—the wire-like paths carrying their signals. Both could produce similar signals, making it difficult to separate a user’s movement from interference, according to MIT’s explanation.

X-Hinges instead prints two conductive filaments with different conductivities inside a body that can bend or compress. One material serves the sensors; the other carries signals while reducing interference. The paper describes embedding the sensing elements during fabrication, without a separate sensor-assembly step afterward.

Users connect shapes in the design tool and choose how each part should move. Options include a cylinder, rectangle, thin outline or custom shape. The software places sensing structures where the object bends. It can also add those structures to existing 3D models, rather than requiring every design to start from scratch.

The researchers report sensing across up to three kinds of movement: compression, vertical bending and lateral bending. Each has a dedicated sensing configuration. A data-acquisition system collects the signals, and data-driven regression models translate them into continuous, real-time motion estimates.

A grasp, a pinch and a shark’s tail

The glove demonstration used blocks to show the human-to-robot transfer. Its design adjusts to the wearer’s hand size. The researchers suggest this approach could help robots learn new interactions in homes or factories; those settings are proposed uses, not deployments described in the account.

Other examples turn deformation into a control input. An origami-style lamp changes brightness when someone pinches it closed. For a shark-shaped game controller, the team drew a custom tail and chose an up-and-down bending motion, allowing it to function like a joystick.

The team also built a device that identifies objects placed on it using tactile sensor data, including a banana. The researchers propose possible detection uses in airports and grocery stores. That example extends the work beyond tracking a joint: the sensed physical interaction becomes information about an object.

Identical shapes can give different readings

Printing the sensors does not eliminate calibration. Small differences in the arrangement of electrical elements mean two objects that look identical can produce different readings. That becomes important when making the same glove for separate users: matching the outer design does not guarantee matching sensor outputs.

The team built a computer-vision system to recalibrate outputs across copies of a device. A built-in signal-recalibration mechanism remains a future goal. Senior author Jiaji Li also wants to combine sensing with printed actuation—the ability to make an object move—in one fabrication process. X-Hinges supplies the sensing side of that proposed combination.