SBIR-STTR Award

Marble associates, Inc., Proposes to utilize the mobile autonomous robot simulator (MARS) system (developed for the united states air force under f33615-87-c-3405) to develop and evaluate various algorithms and heuristics to "fuse" (i.e., compare, contrast, and integrate) the often conflicting data returned by an assortment of non-contact ranging sensors. At present, marble is running a prototype version of the simulator, and we are currently involved in negotiations with the USAF wright aeronautical laboratories for phase ii funding to build a production version based on the detailed design specification produced in the fall of 1987. When implemented, the mars system will permit the autonomous mobile robotics engineers to experiment with a wide variety of heuristics and algorithms for guidance and control of robotic platforms without having to build and maintain an actual robot. The novice user will be able to monitor the performance of various off-the-shelf, non-contact, ranging sensors in unknown, potentially hostile environments, while the expert user will be able to add "device drivers" for new sensors quickly and efficiently,, without having to build a robot around the sensor. In addition, the user will be able to enter man-in-the-loop mode to communicate directly with lower-level modules of the simulator (in effect, changing the platform from a truly autonomous vehicle into a "teleoperated" vehicle). The design specification for the architecture of the simulator Includes an isolated software module known as the "sensor manager" whose sole purpose is to resolve discrepancies among conflicting sensor readings. Marble proposes to develop a full-fledged version of the sensor manager to be used as a tool to aid robotics engineers in sensor evaluation.

During phase ii, the contractor will deepen his understanding of the p-3 domain. In particular and he shall select (in consultation with p-3 flight instructors and the staff at the ai center) a constrained domain of sensor managment. He shall refine and perfect knet according to the specifications outlined in the following chapters. As a test of knet's robustness and effectiveness, and then shall deliver a knowledge-intensive textbook for a specific classification problem that incorporates a deterministic model of acousti propagation. The delivered software will maximally demonstrate knet's integration into a general-purpose computing environment; the incorporation of hypertext and video images; knet's tools for knowledge assessment; its facilities for capturing probabilities and utilities; and its normative foundation in the theory of belief networks and efficient algorithms. The training application will conclusively demonstrate knet's importance not only to the navy, but to industry as well.