This paper describes off-road perception techniques developed for the Autonomous Land Vehicle (ALV). Processing is performed on a parallel processor, the Warp Machine, a 100 MFLOPS systolic array computer. An overview of the complete off-road system is presented, followed by detailed descriptions of the parallel perception processing techniques used in classifying outdoor terrain. Range images are used to determine a Cartesian height map representation of the area in front of the ALV. Then one of a number of different methods of classifying terrain regions in the Cartesian map can be used. The resulting traversability map, consisting of 3 categories (traversable, obstacle, or unknown), is used in path planning. The advantages and limitations of the different approaches are exam- ined and results from each are presented.
Steven B. Seida,
"Parallel Off-Road Perception Processing On The Autonomous Land Vehicle", Proc. SPIE 1007, Mobile Robots III, (10 March 1989); doi: 10.1117/12.949081; https://doi.org/10.1117/12.949081