Abstract
The RoadView project/spl trade/ at the Advanced Highway Maintenance Construction Technology (AHMCT) research center seeks to improve the safety and efficiency of snow removal by providing information to the driver using an in-vehicle computer. The calculation of future vehicle lateral position is achieved with cooperative modular artificial neural networks, trained using data generated from a known, but somewhat slow, mathematical model. The performance of a single monolithic neural network is compared against a cooperative modular neural network trained with uniform matching criteria. Additionally, an algorithm to calculate a best achievable matching criterion for each network is described and the best achievable matching criterion is combined with a modular network partitioning scheme. The use of cooperative modular artificial neural networks reduces mean error between 46% and 55% compared with the single network.