The Foveal Extra-Vehicular Activity Helper-Retriever (FEVAHR) is a mobile robot that features a hierarchical foveal machine-vision system (HFMV). The FEVAHR is a prototype of future robots that could detect, recognize, track, and pursue objects and avoid obstacles while operating autonomously, controlled by human operators via natural-language commands, or both. The design of the FEVAHR merges high- and low-level anthropomorphic designs. The high-level anthropomorphism is represented by (1) the Semantic Network Processing System (SNePS) software for semantic representation of information, inference, and natural-language interaction, and (2) the Grounded Layered Architecture With Integrated Reasoning (GLAIR) software, which acts as an interface between SNePS on the one hand and subconscious processes and sensors on the other hand. The low-level anthropomorphism is represented by the HFMV hardware and software, which exploit the neuromorphic multiacuity sensing and information processing prevalent among vertebrates to achieve an effective visual information-acquisition power that is higher than that of uniform-acuity active vision. SNePS, GLAIR, and HFMV work in unison, each driving and being controlled by the others, to accomplish physical tasks with constrained resources and maintain a high level perception necessary for autonomous interaction with humans.

This work was done by Andrew Izatt, Christopher A. Kramer, Cesar Bandera, and Fenglei Du of Amherst Systems, Inc., and Stu Shapiro and Henry Hexmoor of the State University of New York for Johnson Space Center.