I. Introduction
Teleoperation gives the possibility to remotely control robots in an intuitive way. The human operator expresses his/her intentions by means of a specific input device such as body-worn InterSense IMUs [1], exoskeletons [2], or joysticks [3] and the collected data are used to drive the robotic system. The operator can adapt his/her actions according to the specific task by exploiting his/her experience and creativity [1] [2] [4]. Thus, teleoperation combines the specific skills of the robotic system with human decision capability. For this reason, it can be applied to a wide range of fields such as space [5], underwater [6], nuclear dismantling [7] and Programming by Demonstration [8]. In teleoperated systems, collision avoidance plays a fundamental role since it makes teleoperation safer and easier. The human teleoperator can focus on the task without worrying about possible collisions. Self-collision prevention is generally managed following two main steps: the calculation of distances between the parts of the robot, and intervention once a potential collision is identified. According to the specific implementation of these two phases, many approaches could be distinguished in the literature. To calculate distances between links/end-effector(s) of the robot, their geometries are approximated with different shapes, to meet the desired accuracy and complexity. The main types of approximation are here listed by increasing complexity and accuracy: spheres [9]–[12], cylinders [10] [13]–[16], capsules [17] [18], swept volumes [19], sphere-torus-patches bounding volumes [20]. A high accuracy allows to increase the range of motion of the robot. On the other hand, it might also increase the computational effort, and thereby compromise the real-time capabilities by introducing an excessive delay between the movement of the operator and of the robot. Concerning the type of intervention against possible collisions, the main approaches are two: stopping the robot or planning a new trajectory.