Model-Free Co-Optimization of Manufacturable Sensor Layouts and Deformation Proprioception

A team from The Chinese University of Hong Kong (CUHK), led by Professor Charlie C.L. Wang, has introduced a novel computational pipeline that automates the design of sensor systems for soft robots and wearables. The core innovation is a model-free, data-driven method that co-optimizes two critical elements: the physical layout of flexible, length-measurement sensors and the parameters of a neural network that predicts shape deformation from sensor signals. Unlike traditional approaches that rely on trial-and-error or complex physical simulation models, this pipeline requires only a dataset of deformed shapes, making it broadly applicable to diverse sensing tasks where accurate models are hard to build.

The pipeline uses differentiable loss functions to balance prediction accuracy with real-world manufacturability constraints, ensuring the optimized sensor layouts are practical to fabricate. By jointly training the sensor placement and the prediction model, the system discovers optimal configurations that a human designer might never consider. The researchers validated their method through numerical and physical experiments on multiple soft robotic and wearable systems, demonstrating significantly improved deformation proprioception—the sense of one's own shape and movement—compared to layouts designed heuristically. This breakthrough could accelerate the development of more responsive and intelligent soft robots, advanced prosthetics, and smart clothing.