Customizable Piezo‐Transmittance‐Based Multiaxial Force Sensor With Deep‐Learning‐Assisted Sensing

ABSTRACT The development of Human Machine Interfaces (HMIs) has enabled efficient interactions between robots and operators. Soft sensors are essential for their elasticity and ability to conform to complex surfaces, enabling safe and accurate sensing in dynamic environments. However, many soft force sensors are limited to single‐axis detection or lack the directional accuracy and robustness required for real‐time applications. This work presents a tri‐axis piezo‐transmittance sensor integrated with deep learning for real‐time force direction and intensity prediction in HMI applications. The sensor employs three LEDs and photodiodes with an interlayer of three distinct opaque elastomeric tilted structures that modulate transmitted light according to the direction and magnitude of applied force, enabling tri‐axial detection. Photodiode output voltages are processed by a tailored 1D convolutional neural network (CNN), achieving 99.6% classification accuracy for force direction and a mean squared error (MSE) of 0.0003 N 2 for force intensity prediction. Practical demonstrations include multi‐directional force control of a 2D role‐playing game (RPG) and real‐time grip adjustment and danger detection in a robotic hand.