Deep Learning LSTM-Based Slip Detection for Robotic Grasping

Abstract

Robotic grasping is critical to manufacturing, logistics, and warehouse operations. With the rise of automation and robotization, robots are increasingly being used to pick and place products of various sizes, shapes, and materials in Any-Mixed-Any-Volume scenarios. However, this process can be challenging when dealing with fragile products with mixed volumes, such as glass items and instruments. As a result, slip detection is a key technique used to address these challenges, enabling robots to adjust their grip force and prevent the dropping of fragile products. This research firstly provides an overview of the challenges associated with robotic grasping, particularly in mixed-volume scenarios, and highlights the importance of slip detection in improving efficiency and reducing failures. Next, DIGIT tactile sensors are mounted on the left and right fingertips of the Robotiq 2f-145 gripper to stream the 2-axis tactile images during the grasping motion. Last but not least, a deep learning-based slip detection model is proposed to predict the slipping during the grasping sequence. Experiments show the improvement in success rate with anti-slip detection activated.