Flexible Sensing for Embodied Robotics: From Materials and Devices to System Integration and Closed-Loop Control

Abstract

Flexible sensors that bend, stretch, and conform to curved surfaces are being integrated into robotic bodies with increasing frequency, allowing robots to acquire contact, deformation, and near-field environmental information on curved and large-deformation structures. High-sensitivity devices, however, do not automatically produce reliable task-level perception. This review is organized along a traceable evidence chain from observable, transduction, body integration, signal interpretation, and closed-loop control to task outcome, and compares five transduction mechanisms (piezoresistive, capacitive, piezoelectric, triboelectric, and optical) together with material platforms such as liquid metals and hydrogels. It further examines how structural assembly, stretchable interconnects, calibration and drift compensation, multimodal fusion, and feedback control affect system performance. Using dexterous manipulation, deformation and posture estimation, and human-robot interaction as the three task domains, the review separates device metrics, robot-integration demonstrations, and closed-loop task evidence, and applies this evidence grading consistently across all layers. Current work shows that contact tactile sensing has formed a relatively complete real-robot closed loop under limited objects and operating conditions, whereas environmental sensing, cross-platform transfer, and long-term whole-body deployment still lack unified validation. Future priorities should shift from isolated sensitivity improvement toward comparable evaluation under unified conditions, reporting of end-to-end latency and maintenance costs, and perception-decision-action loops with fail-safe fallback.