Transforming the ego-centered internal representation of an autonomous robot with the cascaded neural network

This paper addresses the problem how the ego-centered internal representation of a robot is to be transformed upon robot movement if the robot´s environment is represented in an occupancy grid. The transformation rules are derived and it is shown that for a single change in the robot´s position, the parameters of this transformation can best be estimated with Monte Carlo sampling. A neural network architecture is introduced as a computational model of the Monte Carlo estimation method, which can calculate estimates of all parameters in parallel. The cascaded neural network is an extension to this architecture, which is capable of learning the relation between the change in the robot´s configuration and the parameters of the corresponding transformation of occupancy grids