Exploring movable space using rhythmical active touch in disordered obstacle environment

We propose a novel navigation system for adaptively exploring an obstacle space using diverse ways of touching an object. Conventional navigation models are typically based on the avoidance of obstacles, i.e., avoiding collision. However, actual disordered space may be full of various kinds of obstacles. To reach a destination in such a space, a robot requires an active approach for avoiding a deadlock with obstacles or changing the obstacle configuration to find an open space using diverse ways of touching an object. We solved this problem by generating locally diverse moving patterns by using an action model with rhythmical oscillation in addition to a localization model using a particle filter. The proposed model was demonstrated to be effective through an experiment where a robot navigated to a destination behind partially movable obstacles using rhythmical active touch.