Touch Based Localization for High-Precision Manufacturing

Abstract

In traditional automated manufacturing, the motions of an industry robot are programmed step by step by humans for a specific task, which is costly and time-consuming. This greatly limits the application of industry robot when the production is relatively small. This thesis proposes two touch-based localization approaches extended from the importance-sampling particle filter for a potentially large and complex object in 3D workspace, both of which can achieve high precision and real-time localization speed and can be potentially used for automated manufacturing. The rigid-body particle filter assumes that an accurate geometry model of the object is provided, while the datum particle filter takes the engineering tolerance for manufactured part into consideration and provides a method to localize a task location defined by datums on an object with internal degrees of freedom.  Both approaches have been evaluated in simulation. Our results show that the proposed approaches can quickly reduce the estimation accuracy in both translational and rotational dimensions within 10 measurements.