Servoed world models as interfaces between robot control systems and sensory data

SUMMARY A hierarchical robot sensory system being developed for industrial robotics is described. At each level of the hicrarchy, sensory intcrpretativc processes are guided by exl~cctancy-enc crating modeling processes. The modeling processes are driven by u priori knowledge (object prototypes). by knowledge of the robot's movements (feed-forward from the control system), and by feedback from thc intcrprctative processes (prior state of the sensory world). At thc lowest level, the senses (vision, proximity, tuctilc, force, joint angle, etc.) are handled separately; above this Icvcl. they are integrated into a multi-model world modcl. At successively higher levels, the inter-prctativc and modeling processes describe the world with successively highcr order constructs, and over succes-sivcly longer time pcriods. Each level of the modeling hierarchy providcs output, in parallel, to guide the cor-rcspcmdinp levels of a hierarchical robot control system.