A Vision-Based Learning Method for Pushing Manipulation

We describe an unsupervised on-line method for learning of manipulative actions that allows a robot to push an object connected to it with a rotational point con- tact to a desired point in image-space. By observing tile results of its actions on tile object's orientation in image- space, the system forms a predictive forward empirical model. This acquired model is used on-line for manipula- tion planning and control as it improves. Rather than ex- plicitly inverting the forward model to achieve trajectory control, a stochastic action selection technique (Moore, 1990) is used to select the most informative and promising actions, thereby integrating active perception and learn- ing by combining on-line improvement, task-directed ex- ploration, and model exploitation. Simulation and exper- imental results of the approach are presented.