Autonomous Discovery of Multi-Material Metamaterials With Programmable Multimodal Thermal Responses via Reinforcement Learning

Inverse design deep learning models have significantly accelerated materials discovery through mapping complex structure-property relationships. However, they remain constrained by prohibitive data demands and are limited to in-distribution interpolation designs. Efficient out-of-distribution inverse design under sparse data constraints remains a critical bottleneck. Here we present an inverse design framework integrating Transformer encoder with reinforcement learning (RL) agent. By combining adaptable structural parameterization and hybrid iterative optimization, this framework efficiently navigates the complex, high-dimensional multi-material design landscape with extremely low data dependency (<10³ samples). The inverse design framework leverages reward-driven mechanism to autonomously evolves multi-material architectures that achieve unseen target thermal deformation behaviors while strictly adhering to manufacturing requirements. Uniquely, the versatility of this framework is demonstrated through the realization of multiple thermal deformation targets, including near-zero, negative, and highly anisotropic thermal expansion, as well as complex multimodal deformations. Experimental validation of fabricated structures further confirms the efficacy of our framework for accelerated on-demand design of multi-material functional architectures.