Machine learning–driven design of engineered cilia enables hybrid operations in acoustic microrobots

Abstract Microrobotic systems offer significant potential for precision medicine by enabling minimally invasive interventions in complex fluidic environments. However, effective operation in these settings requires actuators capable of more than simple linear or rotational motion, often necessitating programmable changes in both direction and shape. This remains a major challenge due to fundamental constraints in the design and control of microscale actuators, particularly in acoustic systems. Here, we introduce engineered cilia for hybrid operations microrobots, a class of acoustic microrobots that use geometry-tuned cilia and resonance-induced forces to execute complex motions such as bidirectional bending, controllable rotation, and adaptive morphing. The microrobots design is driven by a self-augmenting machine learning framework integrated with finite element analysis, enabling rapid prediction and optimization of geometry-resonance relationships across design space. This approach achieves >10⁵-fold reduction in prediction time and over 20-fold in memory savings, while maintaining >90% accuracy in peak amplitude and >98% in resonance frequency. Compliant mechanism strategies further expand the mechanical versatility of the microrobots, enabling programmable shape transformations tailored to specific tasks. These advances establish acoustic-driven microrobots as a scalable and efficient platform for intelligent microrobotic actuation in biomedical and microfluidic applications.