A glove that teaches a robot how to grasp, a lamp that reacts to a pinch, and a platform that recognizes what rests on it share a common feature: each is designed to sense physical interaction from within. X-Hinges, software developed by researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Tianjin University (TJU), helps users design such objects for 3D printing.
Separating sensor signals from electrical noise
These devices use conductive materials to create sensors and wire-like traces inside a printed object. When both parts are made from the same material, they can produce similar signals. The resulting interference makes it harder for a device to distinguish a sensor’s reading from other electrical activity.
That confusion can affect how well an object interprets movement. In an earlier glove example, unclear signals made it difficult to filter noise and confirm the angle of a user’s hand motions, contributing to errors when a humanoid hand tried to copy them.
X-Hinges addresses the problem by assigning different materials to the two jobs. Its software embeds one filament for sensors intended to produce a strong signal and another for traces that carry connections without interfering with those readings. The aim is to help a printed object capture interactions more reliably.
Designing movement into a printed object
Users build a design by joining shapes and choosing how its parts should move. The software supports movement along up to three axes: compression, up and down, or sideways. Sensors can be placed at bending points, such as joints, so the object records movement from more than one direction.
For a robotic hand, for example, fingertips may need to move vertically and laterally. Sensors at the joints can capture both kinds of motion, giving the system more information about how the hand is being used.
The interface offers a custom shape, a cylinder, a rectangle and a thin outline. Researchers used a custom shape to make a cartoon shark controller, setting its tail to move up and down like a joystick. X-Hinges then placed the sensors in the design before the team printed the device.
From robot training to a responsive lamp
One demonstration is a glove that fits a user’s hand and senses how their finger joints bend. It sends that motion information to a humanoid hand, which can imitate a grasp of blocks. The researchers suggest that this kind of glove could help robots learn new interactions quickly in homes or factories.
Another prototype is a small, origami-style lamp. It detects when someone pinches it closed and adjusts its brightness in response. When switched on, the portable light unfolds from a shape resembling folded paper.
The researchers also made a device that identifies objects placed on it using tactile data from its sensors. In one example, it recognizes a banana. The article describes possible uses for this kind of detection in settings such as airports and grocery stores.
Together, the examples show how ordinary objects could respond to touch and movement as part of their intended task. A lamp could react to how it is handled, while a toy or controller could register motion in several directions. Jiaji Li, an MIT electrical engineering and computer science (EECS) postdoc and CSAIL researcher, describes the idea as adding a conscious element to everyday objects.
Calibration remains a challenge
Printed copies of the same design may not give identical sensor readings. Small differences in where electrical elements end up can change a device’s output, even when two objects look alike. That variation matters when comparable items, such as deformable gloves for different users, need to provide consistent data.
To address this, lead author Xiang Chang, a TJU PhD student and visiting researcher at CSAIL, and Li built a computer vision system that recalibrates outputs so readings from two copies can be aligned. The researchers hope to develop a built-in signal recalibration mechanism in the future.
Li’s broader research goal is to build movement and sensing directly into printed objects instead of adding those capabilities afterward. He previously developed tendon-driven mechanisms that let objects move. A next step for the work is combining actuation and sensing in one print, which the researchers say could create new physical platforms for embodied intelligence and AI agents.
Chang and Li wrote the paper with MIT associate professor of EECS and CSAIL principal investigator Stefanie Mueller and recent TJU graduate Haiyang Yan. Their work was supported, in part, by a postdoctoral research fellowship from Zhejiang University and the MIT-GIST Program. The researchers will present their work at the ACM Symposium on User Interface Software and Technology in early November.