Autonomous environment recognition by robotic manipulators

This paper discusses methods of autonomous environment recognition and action by a robotic manipulator working with dynamic interaction to the environment, e.g., assembling. A method automatically recognizes the contacting situation with the work site from the sensor outputs and the robotic manipulator motion. The autonomous recognition then discriminates the constraint conditions at manipulator hand using the self-organizing map that is a kind of unsupervised learning of neural networks. The discrimination of the constraint conditions is successfully demonstrated by a numerical simulation of a 3-link SCARA type manipulator. Another is for the cognitive action. Some approaches based on the reinforcement learning are proposed. They give models of cognitive actions and approaches to so-called frame problem obstructing efficient learning and action.