A MODEL-BASED APPROACH TO CARTESIAN MANIPULATION CONTACT SENSING

This article describes a new statistical, model-based approach to buil ding a contact-state observer for time-invariant constraints for a poi nt in three dimensions. Three-dimensional constraint estimation and se lection, and the application of these procedures to a planar, two-dime nsional maze is described in detail. The observer uses measurements of the contact force and position, and prior information about the task encoded in a network, to determine the current location of the robot i n the task-configuration space. Each node represents what the measurem ents will look like in a small region of configuration space by starin g a predictive, statistical, measurement model. Construction of the ta sk network requires a model of both the grasped part and the environme nt. The model makes the system robust to alignment errors, however, gr oss errors can occur if the topology of the modeled configuration spac e differs from the true topology. Arcs in the task network represent p ossible transitions between models. Beam search is used to match the m easurement history against possible paths through the model network to estimate the most likely paths for the robot. The resulting approach provides a new decision process that can be used as an observer for ev ent-driven manipulation programming.