A mechanically embodied, double-network hydrogel adaptive memory element for reprogrammable actuation

Living things do not just react to the world; they remember it. A muscle that has been exercised responds differently to a new load than one that has rested, and a nervous system learns by altering how it processes future signals based on past experience. This ability to store an internal state that changes how a material behaves is a hallmark of biology, yet most man-made machines lack it. A synthetic robot arm, for instance, is built with a fixed set of instructions: push a button, and it moves a specific distance. It does not learn from the movement, nor does it change its behavior based on how many times it has moved before. Scientists have long sought to build "physical intelligence" into soft materials, creating devices that can sense, decide, and act without a central computer. The challenge has been to make these materials remember their history not just as a temporary echo, but as a stable, rewritable state that can be deliberately changed, much like writing and erasing a note on a piece of paper.

Researchers at the University of Bristol and the Australian National University have developed a soft, water-based material that achieves this kind of mechanical memory. They created a hydrogel actuator, a small rod made of a jelly-like substance that bends when exposed to light. What makes this material unique is that it combines two different behaviors into a single structure. One part of the gel reacts instantly to light, bending quickly and then slowly returning to its original shape, much like a spring that loses its tension over time. The other part of the gel holds onto a memory of the light it has seen, changing its internal stiffness in a way that lasts for days. By linking these two parts together, the researchers built a device that can be trained to bend more or less than usual, and then retrained to return to its original behavior, all while remaining wet and flexible.

The material is constructed from two interwoven networks of molecules. The first network contains gold nanoparticles and a temperature-sensitive polymer. When a laser shines on this part of the gel, the gold particles heat up, causing the polymer to shrink and the rod to bend. This reaction is fast and reversible; once the light is turned off, the gel cools down and slowly relaxes back to its straight shape. If the light is pulsed repeatedly with long breaks in between, the gel returns to its starting position each time. However, if the pulses come quickly, the gel does not have time to fully relax, and the bending accumulates slightly. This behavior is known as a "leaky integrator," where the material briefly remembers recent events but eventually forgets them as the heat dissipates and the water inside the gel redistributes.

To give the gel a permanent memory, the researchers added a second network containing a special molecule called azobenzene. This molecule can switch between two different shapes, or isomers, depending on the color of light it receives. One shape is stable and stiff, while the other is less stable and softer. Crucially, once the molecule switches to the stiff shape, it stays there for a very long time unless hit with a specific color of light to switch it back. When the researchers built their double-network gel, they found that the stiffness of the hinge where the rod bends depended on which shape the azobenzene molecules were in. If the molecules were in the stiff shape, the gel bent more sharply when hit with the laser. If they were in the soft shape, the gel bent less.

The team demonstrated that they could use light to "write" a new behavior into the gel. By shining a blue laser on the hinge of a gel that was initially in a soft state, they forced the molecules to switch to the stiff state. After this training, the gel bent significantly more when tested with a standard laser pulse, even though the test pulse was identical to the ones used before. This increase in bending was not a temporary effect; it persisted for days. The researchers then showed that this memory could be erased. By shining a green laser on the hinge for an hour, they switched the molecules back to their soft state, and the gel returned to its original, less responsive behavior. They could then write the stiff state again, proving that the material could be reprogrammed repeatedly without degrading or needing to dry out.

This work rules out the idea that the change in behavior was caused by simple fatigue or damage to the material. The gel did not just get tired and stop bending; it actively changed how it responded to the same input. It also showed that the memory was not just a fixed shape held in place by drying, as seen in some other "shape-memory" materials. Instead, the memory was stored in the chemical state of the molecules themselves, which altered the mechanical properties of the wet gel while it was operating. The researchers found that the effect was asymmetric: switching the molecules to the stiff state increased the bending, but switching them back did not decrease it in a perfectly symmetrical way. This suggests that the change in stiffness alters how the force is transferred through the thickness of the gel, creating a non-linear response that depends on the history of the material.

The significance of this discovery lies in how it changes the way we think about soft machines. Instead of building a robot that relies on external sensors and computers to decide how to move, this material embodies the decision-making process within its own structure. It can learn by doing, adjusting its response based on what it has experienced, and then unlearn that behavior when the situation changes. The researchers describe this as a "memristive" system, a term borrowed from electronics to describe a component whose resistance depends on its history. In this case, the mechanical resistance of the gel changes based on the light it has absorbed. By coupling a fast, reactive system with a slow, persistent one, the team has created a material that can store information and use it to modify its future actions. This approach offers a new path toward adaptive soft machines that can operate in unpredictable environments, learning from their interactions and adapting their behavior without the need for complex external control systems.