DOF-decoupled active force sensing (D-DAFS): A human-inspired approach to touch-based localisation tasks
Résumé
In the context of touch-based object localisation in structured environments, solving the full 6D problem is computationally expensive as complexity scales exponentially with the number of degrees of freedom. Previous related works provided solutions to problems whose initial uncertainty was bounded by the computational feasibility for on-line applications. Observing the results of an experiment in which human beings face a blind localisation task, a new planning approach, named DOF-Decoupled Active Force Sensing (D-DAFS), is presented for such robotics applications. Specifically, the whole task is divided into a sequence of subtasks, each of them focused on reducing a piece of uncertainty, so that the resolution of the inference model can be increased accordingly. This allows a better allocation of resources, focusing on adopting the most accurate, thus expensive, scheme only when the uncertainty is sufficiently low for on-line calculations. The proposed methodology has been applied to perform a 3-DOF localisation with a Staubli RX90 robot, and a series of simulations have been run to prove its effectiveness in coping with high initial uncertainty avoiding computational overheads.