Topographic design in wearable MXene sensors with in-sensor machine learning for full-body avatar reconstruction

Wearable strain sensors for full-body motion monitoring often lack customizable working windows to match specific joint/muscle deformation ranges, leading to suboptimal performance and data processing challenges. Here, wearable Ti3C2Tx MXene sensor modules with in-sensor machine learning are fabricated for full-body motion classification and avatar reconstruction, operating via wireless streaming or edge computing. Wrinkle-like topographies are heterogeneously engineered on piezoresistive MXene nanolayers through localized thermal contraction to control crack propagation, yielding ultrahigh sensitivities (gauge factor > 1000) within tunable working windows that cover all joint deformation ranges (6–84%). A wireless module integrates the sensors with Bluetooth to stream multi-channel data and train an artificial neural network achieving 100% accuracy for classifying diverse full-body motions. An edge module integrating the sensors with a machine-learning chip runs an in-sensor CNN to reconstruct high-precision avatar animations of continuous full-body motions with an average determination error of 3.5 cm, while consuming 71% less power than wireless streaming and requiring no external computing devices.

Haitao Yang, Jiali Li, Xiao Xiao, Jiahao Wang, Yufei Li, Kerui Li, Zhipeng Li, Haochen Yang, Qian Wang, Jie Yang, John S. Ho, Po-Len Yeh, Koen Mouthaan, Xiaonan Wang, Sahil Shah, Po-Yen Chen