A Data-Driven Approach for Contact Detection, Classification and Reaction in Physical Human-Robot Collaboration

This paper considers a scenario where a robot and a human operator share the\nsame workspace, and the robot is able to both carry out autonomous tasks and\nphysically interact with the human in order to achieve common goals. In this\ncontext, both intentional and accidental contacts between human and robot might\noccur due to the complexity of tasks and environment, to the uncertainty of\nhuman behavior, and to the typical lack of awareness of each other actions.\nHere, a two stage strategy based on Recurrent Neural Networks (RNNs) is\ndesigned to detect intentional and accidental contacts: the occurrence of a\ncontact with the human is detected at the first stage, while the classification\nbetween intentional and accidental is performed at the second stage. An\nadmittance control strategy or an evasive action is then performed by the\nrobot, respectively. The approach also works in the case the robot\nsimultaneously interacts with the human and the environment, where the\ninteraction wrench of the latter is modeled via Gaussian Mixture Models (GMMs).\nControl Barrier Functions (CBFs) are included, at the control level, to\nguarantee the satisfaction of robot and task constraints while performing the\nproper interaction strategy. The approach has been validated on a real setup\ncomposed of a Kinova Jaco2 robot.\n