AI-Enabled Soft Versatile Grasping for High-Mixed-Low-Volume Applications with Tactile Feedbacks

Abstract

Handling mass customized products is one of the key challenges in manufacturing and logistic industries. Flexible solutions which can cater to novel objects are desirable in these industries which are continually producing unique catalogues of products. However, most of the robotic grasping solutions on the market are not suitable for novel objects in high mix and low volume scenarios. Currently, the gaps in the areas of grasping accuracy and speed are impeding the widespread adoption of robotic grasping in these industries. This research aims to improve the grasping capability for novel objects and demonstrate robotic grasping using soft grippers based on data driven learning to accommodate novel objects with varying shapes and textures.We compare data driven approach with deep reinforcement learning (DRL) approach and found that the limitations of DRL such as being data-intensive, complex, and collision-prone reduce its industry readiness level. Therefore, we opt for PointNetGPD which is a data driven approach in this research. We have also performed a comprehensive market survey on tactile sensors and soft grippers with consideration of factors such as price, sensitivity, simplicity and modularity. Based on our evaluation, we choose Rochu two-fingered soft gripper with our customized Force Sensing Resistors (FSR) force sensors mounted on the fingertips because of Rochu’s modularity and compatibility with these tactile sensors.Finally, we conduct data training using soft gripper configuration and test various fast-moving consumer goods (FMCG) products inclusive of fruits and vegetables which are unknown to the database. The grasping accuracy is improved from 75% based on Push and Grasp to 81% based on PointNetGPD. Our versatile grasping platform is independent of gripper configurations and robot models.