Simulation, Design, and Application of Intelligent-Edge-Based Soft Magnetic Tactile Sensor With Super-Resolution

Abstract

Tactile is one of the sensation foundations for robots to achieve dexterous manipulations and trusted interactions. Among the proposed tactile sensor solutions in literature, soft magnetic tactile sensors have received widespread attentions due to their advantages, such as replaceable elastomers, high frequency, high sensitivity, and super-resolution (SR) potentials. In traditional sensor architectures, the sensors collect raw sensing data, which are transmitted to the PCs for the SR algorithms for the feedback control of actuators later on. Therefore, there is an irreconcilable contradiction between the large amount of data processing for high resolution and the real-time requirements for the control of actuators. In this article, we have designed an improved soft magnetic tactile sensor. Its elastomer thickness, magnetic particles’ doping ratio, and the sensitive element layout are optimized based on a simplified theoretical model. An intelligent tactile sensor is achieved by performing SR model reasoning independently with quantized convolutional neural network (CNN) model at the edge, saving the trouble of great data transmission between sensor and PC. An average cycle time is $3260~\mu {s}$ for each edge-based inference. The RMSE of the contact position and force estimation reaches 0.2689 mm and 36.24 mN, respectively. Meanwhile, the wireless connection among intelligent edge sensors via Bluetooth/Wi-Fi enables free displacement of the sensors at various locations of robots in single, pair, or matrix form for various real-time sensory feedback applications with high resolution, which are also demonstrated in this work. This work would provide reference for the design and implementation of intelligent edge sensors of robots.