Trainable soft electronics with memory in liquid crystal polymers.

State-of-the-art soft materials can be engineered as sensors and actuators, yet, methods for learning from external information remain a subject of current research. Inspired by the use of large datasets to train artificial intelligence, tuning physical responsiveness to relayed data would introduce learning behavior in soft materials. In this work, we develop a trainable liquid crystal oligomer network (LCON) that stores digital information directly into its molecular configuration. By functionalizing the anisotropic LCON with photo-switchable azobenzene, we simultaneously integrate basic logic and memory in a material through a binary-state system; we coin this design the trainable self-propelled gate (T-SPG). We can tune the memory of our T-SPG with photonic stimuli, allowing the system to be trained by a conventional digital controller. We demonstrate the trainability of the T-SPG through two hierarchical tasks: a lower-level binary classification task where the decision boundary is stored as material memory, and a higher-level motion task that uses the stored memory to trigger actuation.