Development of a Parametric Model for the Environment of a Mobile Robot

We present a new approach for reducing the computational complexity of grid-based localization methods by suggesting the development of a parametric model of the environment of a mobile robot. This model is based on the sensor perception and obtained by applying the nonparametric bootstrapping technique. Particle filter based localization methods assume that map is available but are computationally very expensive mainly due to the measurement models. This method reduces the computational complexity of these methods by extracting a parametric model from the given map to be used for measurement update. We assign multinomial probabilities to the grid cells for future inferences. Having the cell probabilities, we can update the particles obtained from action model. The likelihood calculations become straight-forward and hence complex measurement models can be simplified. Our method can exploit any feature extraction algorithm to get the multinomial approximation of the environment. The offline extraction of few vital geometrical feature is carried out at some known locations on the map. This is then followed by the statistical technique to get the model parameters and essential cell statistics. The extracted model is used for measurement updates using a simulated pioneer robot and results show a significant increase in the update rate of the particle filter