A Machine Learning Approach for Visionless Determination of Safe Release Time in Robot to Human Object Handover

A visionless classifier has been developed in order to enable a robot to decide when to release an object in a robot to human object hand over task. The robot should release the object when it is stably grasped by the human and it is safe to be released. Decision-making is based on the forces applied to servo motors in robot joints. The system determines whether the robot should open its gripper to release an object or retain it. Multilayered perceptron (MLP) and support vector machines (SVM) algorithms were utilized to train and test the model for force classification. Utilizing a collected dataset, the algorithms facilitated the development of a classifier capable of deciding whether to release an object. Experimental evaluations, where a visionless robot releases an object and hands it to a human, confirm that the machine learning system can effectively generate a model than makes correct decisions only based on applied forces.