Modelling of Second Order Polynomial Surface Contacts for Programming by Human Demonstration

摘要

This paper presents a contribution to automatic model building in quadratic polynomial environments, in the context of programming by human demonstration. A human operator moves a demonstration tool equipped with a probe, in contact with an unknown environment. The motion of the demonstration tool is sensed with a 3D camera, and the interaction with the environment is sensed with a force/torque sensor. Both measurements are uncertain, and do not give direct information about the different objects in the environment (such as cylinders, spheres, planes,...) and their geometric parameters. This paper uses a Bayesian Sequential Monte Carlo method or particle filter, to recognize the different discrete objects that form the environment, and simultaneously estimate the continuous geometric parameters of these different quadratic polynomial objects. The result is a complete geometric model of an environment, with different quadratic polynomial objects at its building blocks. The approach has been verified using real world experimental data, in which it is able to recognize three different unknown quadratic polynomial objects, and estimate their geometric parameters.