Tactile sensing is crucial for robots exploring unknown spaces, especially in non-visual environments. While existing tactile sensors often limit themselves to single-directional force detection, hindering their application in complex scenarios, the demand for multidirectional tactile perception is high. Current technologies like vision-based sensors (Insight, OmniTact, Geltip) or Hall element-based sensors (uSkin) either necessitate complex structures and algorithms or lack flexibility. Flexible sensors based on Faraday's law of electromagnetic induction offer a promising solution due to their self-generated signal, high sensitivity, and simple structure. Although some self-generated sensors have been proposed, limitations remain in terms of multidirectional sensing and integration into robotic systems. This work addresses these gaps by introducing a novel soft magnetoelectric finger (SMF) employing a liquid metal (LM) coil array for self-generated and multidirectional tactile sensing.

The paper reviews existing tactile sensors for robotic fingers, highlighting limitations in multidirectional sensing capabilities. It discusses existing technologies such as optical-fiber-based sensors, Biotac sensors, and bimodal soft electronic skin, noting their inability to detect multidirectional forces effectively. Vision-based approaches, while enabling multidirectional perception, require complex structures and image processing algorithms. Hall element-based sensors offer multidirectional sensing but lack flexibility. The paper then examines previous attempts at creating self-generated signal sensors based on Faraday's law, noting limitations in achieving multidirectional sensing and soft, flexible designs.

The SMF was fabricated using a two-part process. First, a 3D-printed mold was used to create a flexible Ecoflex sheath with five embedded microchannels. Liquid metal (Ga-In alloy) was then injected into these channels to form five LM coils. Copper wires were embedded for electrical connections, and the assembly was encased in Ecoflex for protection. A 3D-printed phalangeal bone with an embedded NdFeB magnet formed the second part. The SMF's sensing mechanism relies on the change in magnetic flux through the LM coils caused by external forces, resulting in induced voltages. The five coils were arranged to detect forces from various directions. Abaqus and Maxwell software were used for finite element mechanical and magnetic simulations to analyze the deformation and magnetic field changes under pressure. Experiments were conducted to evaluate the SMF's sensitivity to forces of varying magnitudes and directions. A dynamic press setup was used to apply controlled forces, and the resulting voltages were measured using a LinkZill data acquisition system. The data was analyzed to determine the SMF's sensitivity and ability to distinguish different force directions. Machine learning algorithms (LightGBM) were used to classify objects based on the SMF's response to different materials. Finally, the SMF was integrated with a robotic arm to perform black box exploration experiments to evaluate its performance in a real-world scenario.

The SMF successfully achieved self-generated signal and multidirectional tactile sensing. Experiments and simulations demonstrated that the voltage signals generated by the LM coils were directly related to the magnitude and direction of the applied force. The SMF exhibited high sensitivity, detecting forces as low as 0.04 N. Analysis of the signals from the five LM coils allowed for the differentiation of forces applied from different directions, including normal and shear forces. The SMF accurately identified six common objects with different Young's moduli (sponge cubes, porous plastics, paperboard, foamed plastics, Ecoflex rubber, and cured resin) using machine learning, achieving an accuracy rate of 97.46% with a LightGBM classifier. The t-SNE algorithm visualized distinct clustering patterns of the signals from different materials, confirming the effectiveness of the sensor. In black box exploration experiments, the SMF successfully identified obstacles and materials within a confined space, demonstrating its potential for applications in complex, non-visual environments.

The results demonstrate the feasibility of using a self-generated signal soft magnetoelectric tactile sensor for multidirectional force sensing and object recognition in non-visual environments. The successful integration of the SMF with a robotic arm for black box exploration highlights its potential for real-world applications. The high accuracy of object recognition using machine learning suggests that this technology could be used for various robotic manipulation tasks. The work contributes significantly to the field of robotic tactile sensing by offering a solution that addresses the limitations of existing sensors, improving robot dexterity and autonomy. Future research directions include optimizing the design for increased sensitivity, extending the number of sensing units for enhanced directional resolution, and exploring more sophisticated machine learning algorithms to improve the accuracy and robustness of object recognition.

This study presents a novel soft magnetoelectric finger (SMF) that overcomes the limitations of existing tactile sensors by enabling self-generated and multidirectional tactile sensing. The SMF demonstrates high accuracy in object recognition and successful application in a black box exploration scenario. Future work should focus on enhancing sensitivity, improving directional resolution by increasing the number of LM coils, and exploring advanced data processing techniques to broaden its applications.

The current SMF design requires a known compression speed for accurate force determination. While this is manageable in controlled robotic applications, it represents a limitation in more dynamic scenarios. The limited number of LM coils restricts the precision of directional sensing. The current object recognition capabilities primarily rely on differences in Young's modulus, limiting its versatility. Further research is needed to overcome these limitations and improve the robustness and cost-effectiveness of the SMF.