Model-Based Object Pose Refinement For Terrestrial and Space Autonomy

Abstract

Model-based object pose re nement algorithms have been applied to rack stacking and pallet loading/unloading in the context of automated forklift operations in a warehouse environment. These model-based pose re nement algorithms enable high-precision alignment by utilizing known geometric object models and their salient straight line edges in matching 3-D graphic models to actual video images. An analysis of pose error covariance using an incremental least-squares update technique has been performed to examine pose estimate precision, and a comparison of pose estimates to manual measurements has allowed a quanti cation of absolute accuracy. The algorithms implemented have actually been incorporated into a CMU/NREC facility with successful demonstrations of rack stacking and pallet loading/unloading operations. The pose re nement algorithms implemented have also been successfully tested for Orbital Replacement Unit (ORU) module insertion, and these same algorithms could also be applied to such space applications as autonomous space assembly and various stages of sample return.