To apply BDI agents to real-world scenarios, the reality-gap, between the low-level data (perceptions) and their high-level representation (beliefs), must be bridged. This is usually achieved by a manual mapping. There are two problems with this solution: (i) if the environment changes, the mapping has to be changed as well (by the developer); (ii) part of the mapping might end up being implemented at the agent level increasing the code complexity and reducing its generality. In this paper, we present a general approach to automate the mapping between low-level data and high-level beliefs through the use of transducers. These transducers gather information from the environment and map them to high-level beliefs according to formal temporal specifications. We present our technique and we show its applicability through a case study involving the remote inspection of a nuclear plant.