Discriminating Soft Actuators’ Thermal Stimuli and Mechanical Deformation by Hydrogel Sensors and Machine Learning

Multimodal sensing is crucial for soft robots’ environmental interaction and closed-loop control. The sensing signals of mechanical deformation and temperature changes are often hard to differentiate manually. Herein, two hydrogel sensors are used for simultaneously proprioceptive, thermoceptive, and mechanoreceptive sensing. These stretchable sensors show high sensitivity to strain and temperature changes. Then, a machine learning model composed of a 1D convolutional neural network and a feed-forward neural network is utilized to decode the sensing signal for various stimuli identification. It is demonstrated that the proposed method can accurately predict the soft actuator's body posture changes, such as bending, twisting, and stretching. In addition, the model can discern contact events with or without thermal stimuli. This data-driven method for multimodal sensing discrimination might pave the way for future intelligent soft robots.