Correspondence:
[*]
Corresponding author: Roger Bostelman, Intelligent Systems Division, National Institute of Standards and Technology, 100 Bureau Drive, Stop 8230, Gaithersburg, MD 20899, USA. Tel.: +1 301 975 3426; Fax: +1 301 990 9688; E-mail: [email protected].
Abstract: The National Institute of Standards and Technology's (NIST) Intelligent Systems Division (ISD) is a participant in the Defense Advanced Research Projects Agency (DARPA) LAGR (Learning Applied to Ground Robots) Program. The NIST team's objective for the LAGR Program is to embed learning algorithms into the modules that make up the 4D/RCS (Four Dimensional/Real-Time Control System), the standard reference model architecture that ISD has applied to many intelligent systems. This enables the vehicle to learn to navigate in complex, off-road terrain. The vehicle learns in several ways. These include learning by example, learning by experience, and learning how to optimize traversal. Learning takes place in the sensory processing, world modeling, and behavior generation parts of the control system. This paper describes the 4D/RCS structure, its application to the LAGR program, and the learning and mobility control methods used by the NIST team's vehicle. Results are shown from the series of tests conducted by an independent evaluation team, and the performance of one of the learning algorithms is evaluated.
