Reactive collision avoidance for safer human-robot interaction

Related to human-robot interaction, safety is not a simple attribute encompassed by the concept of dependability: it is a central requirement for the risks occurring during the motion of articulated robot manipulators. Both industrial and service robots need strategies for improving safety of people in their workspace. Collision avoidance algorithms are therefore crucial, and reactive techniques are important for unstructured domains. The so-named “skeleton algorithm” is an effective tool for reactive collision avoidance in industrial and service robotics. It needs good sensory information in order to plan the avoidance trajectories before possible collisions. This is achieved simply for self-collision avoidance (where highly dependable sensory data are available), while it needs further attention, related to safety, when noisy and, possibly, delayed exteroceptive sensory data are used, e.g., for people avoidance by heavy robots. Experiments with robots of these two categories are considered.