Experimental Verification of a Completely Soft Gripper for Grasping and Classifying Beam Members in Truss Structures

Abstract

Robotic object exploration and identification methods to date have attempted to mimic human Exploratory Procedures (EPs) using complex, rigid robotic hands with multifaceted sensory suites. For applications where the target objects may have different or unknown cross-sectional shapes and sizes (e.g. beam members in truss structures), rigid grippers are not a good option as they are unable to adapt to the target objects. This may make it very difficult to recognise the shape and size of a beam member and the approaching angles which would result in a secure grasp. To best meet the requirements of adaptability and compliancy, a soft robotic gripper with simple exteroceptive force sensors has been designed. This paper experimentally verifies the gripper design by assessing its performance in grasping and adapting to a variety of target beam members in a truss structure. The sensor arrangement is also assessed by verifying that sufficient data is extracted during a grasp to recognise the approaching angle of the gripper. Firstly, the gripper is used to grasp each beam member from various angles of approach and readings from the force sensors are collected. Secondly, the collected sensor data is used to train and then test a range of commonly used classifiers for classification of the angle of approach. Thirdly, the classification results are analysed. Through this process, it is found that the gripper is proficient in grasping the variety of target beam members. Despite the uncertainty in the gripper pose, the sensor data collected from the soft gripper during a grasp is sufficient for classification of the angles of approach.