Fusing dynamic percepts and symbols in cognitive systems

A novel method for Bayesian tracking and generalized sensor fusion is proposed. The new approach is based on a localized histogram-like representation, the channel representation. The channel representation allows fusing continuous percepts and symbolic input, which enables the proposed scheme to implement perception-action loops with an additional control input. By connecting several such loops through action-control links, a hierarchical network with flexible structure is generated. The novel tracking loops are validated in a simple experiment with real world data.