2A1-L03 Responsive Grasp : Generating Grasping Behavior through Probabilistic Networks as a Knowledge Model

Abstract

In order that robots coexist and work with human in actual environments, a robot needs to handle a task immediately whenever human asks it. At the thought of the task we want robots to do, quick grasp is one of important tasks. Therefore we propose "responsive grasp," that is generated responsively through relationship between sensors and motors. The responsive grasp consists of learning of probabilistic networks as knowledge model, and generating grasping behavior through the knowledge model. In addition, we classify and abstract learning objects in order to cope with unknown objects. We constructed the responsive grasping system, and verified its feasibility by simulation.