In this paper, we investigate in how far nonlinear dimensionality reduction (DR) techniques can be utilized to tackle particular challenges of sensor data from smart home environments. Smart homes often contain a large number of sensors of various types, providing output in real time, which results in a sequence of high-dimensional, heterogeneous data vectors. We propose that DR techniques can provide a truthful low-dimensional representation (i.e. a compression) of this kind of data, together with a corresponding reconstruction (i.e. decompression). This yields an automatic fusion of uncoordinated raw sensor signals, as well as an economical storage format, with a certain robustness against sensor failure. In proof-of-concept experiments, we present first empirical results to test our approach based on real-world data.