DAS LERNFAHRZEUG NEURAL NETWORK APPLICATION FOR AUTONOMOUS MOBILE ROBOTS

Using Neural Network Technology for information representation and information processing opens a wide and promising field of application in the control area for mobile robot vehicles and for driving support systems. Features like supervised learning and self-learning during the training phase (LERNPHASE) and associations combined with very short response times in the operation phase (KANNPHASE) improve the chance of realizing a high degree of autonomous cooperation with the unexpected. Three steps of investigating Neural Network application for the LERNFAHRZEUG (automatic guided vehicle) using the bottom-up approach will be presented: 1 Sensor/Motor associations with trained obstacle primitives 2 Complex driving ground behavior 3 Dynamic cooperative behavior. Using a developed application simulation system the process of engineering Neural Networks and improving their performances will be discussed including paradigms and architectural evaluations. Future technical real time tasks of automatic guided vehicles in real environments consisting of environment space, event space, system space, reaction space and task space will demand information processing techniques with high tolerance and the ability to handle complex cybernetical situations and processing tasks. Numerical processing techniques with bottleneck effects and programming problems in cooperation and interaction are insufficient. In addition, isomorphic representations in form of transfer chains from the physical situation using adequate sensor configurations, preprocessing techniques, input vectors, network topologies, paradigms, and output vectors for locomotion are important for high performances in autonomy. The background of the LERNFAHRZEUG (autonomous mobile robot) was to create a simulation system for Neural Network integration and testing in application fields with a high degree of relevance to realistic situations. Using a sensor system for object contour lines and a locomotion set, various neural network types were applied and investigated.