IJRR发表，软体机器人传感系统新突破！PneuGelSight 借机器视觉实现高精度本体与触觉感知

Core Insights - The article discusses the development of a soft robotic sensor system called PneuGelSight, which integrates visual and tactile sensing capabilities to enhance the performance of soft robots in industrial applications [1][3]. Group 1: PneuGelSight Overview - PneuGelSight is a soft robotic finger that incorporates a camera and lighting system, enabling high-precision proprioception and tactile sensing during grasping tasks [4][6]. - The design features a 3D-printed corrugated structure for easy extension under pneumatic drive, with a thicker silicone layer on the inside to facilitate bending during inflation [4][6]. Group 2: Optical Design and Functionality - The optical sensing structure consists of an embedded camera, a lighting system, and a soft reflective surface, allowing for clear image capture regardless of finger bending [6][10]. - The system uses color variance in reflected light to calculate surface normals and reconstruct the 3D geometry of objects, enhancing the robot's interaction capabilities [10][11]. Group 3: Performance and Testing - The finger measures 110 mm in length and has a semicircular cross-section with a diameter of 55 mm, manufactured using SLA 3D printing technology [7][9]. - Testing showed that the reconstruction accuracy of the 3D point cloud model varied between 2.12 mm and 8.76 mm across different deformation scenarios, with an average of 5.35 mm [16][17]. Group 4: Tactile Sensing Capabilities - Tactile sensing is achieved by detecting changes in the surface normals of the silicone layer when in contact with objects, allowing for high-resolution reconstruction of surface features [19][21]. - The system can detect forces as light as 0.2 N, comparable to lifting a sheet of A4 paper, and can adapt sensitivity by adjusting the silicone hardness [25]. Group 5: Practical Applications - Demonstration experiments involved a soft robotic gripper exploring an avocado, successfully constructing a 3D model of the object through repeated contact and analysis of surface textures [26][27]. - The results indicated a high fidelity in the reconstructed shape and surface texture, showcasing the technology's potential in object recognition and interaction tasks [27].