Catalogue of Artificial Intelligence Techniques
Author(s): Martin Westhead , Ashley Walker
A model-based system is the standard (or `classical') way of building intelligent robot controllers. They are so-called because at the heart of the control system lies a representation or model of the task space (e.g., in a model-based navigation control system the representation is a map of the environment). Classical models, because they grew out of the example of a control engineer's plant model, generally contain very rich, accurate and detailed descriptions. Typically a model-based system will consist of a single monolithic controller which carries out what is called a sense-think-act cycle. In the first stage all the sensor data is read and compared with the model in order to assess the current state of the world. (Sophisticated systems may build their own models from sensor data, however it is more common for the model to be built by the programmer and given to the robot a priori.) Using symbolic planning techniques the control system then reasons about what actions must be performed in order to achieve its goals. The final part of this cycle involves actually carrying out the actions.
- Newell, A., The Knowledge Level Artificial Intelligence 18 (1982), 87--127.
- Nitao, J.J. and Parodi, A.M., A real-time reflexive polot for an autonomous land vehicle, Systems, man and cybernetics: proceedings of the 1985 IEEE international conference, IEEE, New York, 1985, pp.296--299.