An Adaptive Model for Vision-Based Localisation

We introduce a neuro fuzzy model for localising a mobile robot based on raw vision data Not only does the ap proach involve adaptive fuzzy control but it also integrates data compression such as the principal component analysis PCA as well as pattern matching in eigenspace A typical nonlinear real world system may be modelled as a control process with a large number of sensor inputs By observ ing the control process in a continuous time interval high dimensional sensor data and eventually some desired control information can be sampled With the help of PCA the orig inal dimension of input data can be reduced to a tractable size Subsequently the data can be clustered into a nite sequence of local scenarios Within each local scenario an adaptive fuzzy controller is trained to achieve the desired performance