In a groundbreaking advancement set to revolutionize materials science and robotics, researchers at the University of Amsterdam have engineered metamaterials capable of autonomous learning and adaptation, mimicking traits historically reserved for living organisms. Published in Nature Physics, this pioneering research elucidates how artificially designed metamaterials can dynamically reshape themselves, not only by recalling previously learned forms but also by autonomously optimizing their shape-changing strategies in real time without centralized control. These materials exhibit emergent behaviors akin to reflexes and locomotion, marking a dramatic leap from traditional, passive materials or pre-programmed robotic systems.
Conventional materials respond predictably under mechanical stresses. A steel beam bends a certain way under load; rubber stretches and recoil with fixed elasticity. Robots, conversely, perform tasks through rigid, pre-written codes which dictate their every movement. The innovation from the University of Amsterdam transcends both these realms by drawing on biological inspirations — where cells and simple life forms adapt fluidly to unpredictable environments without any brain or central nervous system. These smart metamaterials embody this property, featuring decentralized computing and learning capacity embedded within their structure, thereby permitting them to evolve their forms autonomously.
The metamaterial crafted by the research team is essentially a chain of motorized hinges interconnected by an elastic core. Each hinge is equipped with a microcontroller that continuously gauges its angular position, stores historical motion data, and communicates with adjacent hinges. Armed with this localized information, individual hinges recalibrate their torque outputs, effectively modifying their stiffness and preferred articulation angles. This networked, cooperative adaptation culminates in the material learning new configurations, gradually mastering complex shape transformations by training on example patterns.
Remarkably, these worm-like metamaterials demonstrate the ability to “forget” outdated forms and embrace novel shapes or, alternatively, to retain multiple form memories that they can toggle between easily. This plasticity enables functionality beyond ordinary static shape changes, allowing the material to interact actively with its surroundings. Their capabilities currently include gripping objects and locomotion — akin to wriggling or crawling motions — that they perform without external directives, underscoring autonomous behavioral adaptation.
The depth of this learning process bears a resemblance to neural mechanisms, but intriguingly unfolds entirely through mechanical and microcontroller synergy without classical centralized intelligence. As Yao Du, the PhD candidate spearheading the study, explains, “Once the system starts to learn, the landscape of possible outcomes becomes nearly infinite. The capacity for evolution-like progression within these materials opens profound opportunities for dynamic, multifunctional devices.”
This research extends prior work from the Machine Materials Lab that demonstrated brainless locomotion in odd-shaped metamaterials by harnessing intricate mechanical design and environmental interactions. The previous iterations facilitated autonomous rolling and crawling over uneven terrain but lacked any learning or memory components. The current leap in enabling adaptive learning transforms these constructs from passive responders into active agents capable of rewriting their own behavioral codes.
Looking ahead, the team aims to transcend static shape transformations toward mastering dynamic, time-dependent behaviors, such as learning various locomotion gaits suited to distinct environmental stimuli. This advancement would empower metamaterials to alter modes — crawling, rolling, or wriggling — based on context, much like living organisms. Additionally, the researchers plan to tackle stochastic learning scenarios where adaptation occurs in the presence of noise and uncertainty, endowing systems with probabilistic learning rather than deterministic responses, which is essential for robustness in wildly fluctuating or unpredictable settings.
This exciting frontier aligns perfectly with contemporary scientific and technological themes. The Dutch Research Agenda’s 2026 call emphasizes “Materials that learn and learning how we responsibly use them,” highlighting the societal and ethical dimensions alongside the scientific breakthroughs. To further these ambitions, a new PhD candidate will join the Machine Materials Lab in close collaboration with the Learning Machines group at AMOLF. This consortium aims to refine and expand upon these transformative materials, ultimately creating smart systems that seamlessly integrate learning at the material level.
What makes this metamaterial paradigm particularly enthralling is its decentralization and scalability. Unlike robotic systems relying heavily on centralized processors and complex algorithms, these metamaterials distribute computation and memory throughout their physical form. Each hinge processes and stores only localized information yet collectively manifests sophisticated learning and adaptation, heralding a new design philosophy where intelligence is sculpted into the fabric itself.
Beyond potential applications in soft robotics and adaptive structures, such materials hold promise for a plethora of industries — from medical devices that conform and evolve to patient anatomy, to aerospace components that self-repair and optimize under stress, and even in wearable technology that adapts to the wearer’s movements and needs in real time. The blending of mechanical ingenuity with embedded, distributed learning algorithms sets a visionary course for future technologies.
The journey from passive materials to active, learning substrates may very well mark a paradigm shift in how humans interface with their environment. These metamaterials imbue physical substance with a semblance of agency — a hallmark of biological entities yet replicated here in engineered matter. As the team’s work progresses, it may pioneer not only smarter machines but also reshape fundamental notions of materiality and interaction.
In summary, the evolution of metamaterials that learn to adapt, remember, and autonomously optimize their functionality may redefine the boundaries between the animate and inanimate. The research from the University of Amsterdam represents a significant stride towards materials that are not static objects but living, learning entities capable of real-time adaptation and multifunctionality.
Subject of Research: Not applicable
Article Title: Metamaterials that learn to change shape
News Publication Date: 7-Apr-2026
Web References:
10.1038/s41567-026-03226-2
Image Credits: Yao Du et al.
Keywords
Physical sciences, Materials science, Material properties, Materials engineering
