Abstract
This paper describes an expert system for objective rating of subjective characteristics like driving comfort. The system uses radial basis neural networks that can be trained on any dynamic properties, for example acceleration. Training and retraining can be done in real time. The system includes a measure of the reliability of automatic judgement, which can be used to signal when new training may be necessary. The paper shows in detail how the system has been used to automatically judge gearshift comfort for automatic transmission. Tests indicate that the system's accuracy and consistency are as good as one of Volvo's best experts. (A)