Design and Integration of an Underactuated Robotic Finger with Vision-based Tactile Sensing

Abstract

Underactuated fingers are adaptable to different shapes, robust, and cost-effective for executing sturdy and versatile grasps. However, they generally have limited control or require complex planning when performing tasks that require high precision or delicate handling. Vision-based tactile sensors, like GelSight, can mitigate these control issues by adding real-time proprioception and also provide useful high-resolution tactile information, which can enhance underactuated fingers with shape and texture perception. As such, this work presents the development of a compact, underactuated linkage finger and its integration with a low-cost, simple vision-based tactile sensor, i.e. the Gelsight.

Through the process of developing the tactile, linkage fingers, we established a planar linkage mechanism simulator and a simple 2D ray-tracing optical simulator to help optimize the linkage transmission and improve tactile sensing performance. In total, the finger went through three major designs. In the initial iteration, we designed sliding joints, which were replaced in the second iteration by linkage mechanisms to make the design more compact and robust. A planar linkage simulator was used to optimize the trajectory to avoid collision and increase range of motion. In the current iteration, the finger has evolved from having two segments to having three segments, with underactuation incorporated to further reduce the number of motors. Each finger segment houses a silicone gel pad, whose tactile imprints are captured by mirrors, which are then observed by a single camera placed at the second finger segment. The camera and mirrors are positioned based on the results of a simple raytracing simulator, which guaranteed that each finger segment could be visible in all finger configurations.

The use of mirrors, linkage transmission and underactuation makes the mechanism compact, efficient, and less complex by reducing the number of cameras and motors needed. Moreover, the integration of vision-based sensors allows these underactuated fingers to perceive contact information and finger configuration. In conclusion, this work encapsulates the innovative design and integration of an underactuated linkage finger with vision-based tactile sensing, offering compactness, adaptability, and robustness in grasping tasks. Additionally, the integration of vision-based tactile sensors can significantly enhance the capabilities of underactuated fingers by providing them with high resolution images and proprioception information, and potentially broaden the future usage of underactuated fingers.