News

Soft robots that remember

Published on November 5, 2025

Alberto Comoretto and his colleagues in the Soft Robotic Matter group at AMOLF have created soft robots that can remember past events – without using computers or electronics. Instead, their memory is built into the robot’s flexible body. The researchers published their paper in the Cell press journal Device, and the robot was even featured on the cover of this journal.

Soft robotic design

The team, led by Bas Overvelde, designed machines powered by air and made of soft materials. At their core is a small elastic shell that can snap between two shapes, allowing the robot to ‘store’ information. When touched or bumped, the shell changes shape, altering how air flows inside the robot and how it moves.

By adding soft antennae that kink when they touch something, the researchers enabled the robots to sense and respond to their surroundings. For example, when one of these robots hits a wall, it changes direction on its own – remembering the obstacle for next time.

This soft fluidic robot without electronics detects obstacles placed on its course and “remembers” their presence by steering away, thanks to the mechanical memory embodied in its structure.

Inspiration

Bas reflects on how it all started: “We were fascinated by the mechanics of the popular popping dome toy –  how it stores energy in one shape and releases it when it snaps. That simple idea became the foundation for our robots’ memory: information encoded directly in their structure and shape, not in electronics.”

These designs could lead to autonomous materials and machines that are simple, robust, and capable of exploring environments where traditional robots cannot go.

Learn more

If you have questions about this research, please contact Bas Overvelde (email: b.overvelde@amolf.nl).

This paper was published in Device by Cell Press: Embodying mechano-fluidic memory in soft machines to program behaviors upon interactions by Alberto Comoretto, Stijn Koppen, Tanaya Mandke, Bas Overvelde.

Read the full paper