Extracting Planar Kinematic Models Using Interactive Perception

Interactive perception augments the process of perception with physical interactions. By adding interactions into the perceptual process, manipulating the environment becomes part of the effort to learn task-relevant information, leading to more reliable task execution. Interactions include obstruction removal, object repositioning, and object manipulation. In this paper, we show how to extract kinematic properties from novel objects. Many objects in human environments, such as doors, drawers, and hand tools, contain inherent kinematic degrees of freedom. Knowledge of these degrees of freedom is required to use the objects in their intended manner. We demonstrate how a simple algorithm enables the construction of kinematic models for such objects, resulting in knowledge necessary for the correct operation of those objects. The simplicity of the framework and its effectiveness, demonstrated in our experimental results, indicate that interactive perception is a promising perceptual paradigm for autonomous mobile manipulation.