Most current approaches to scene understanding lack the capability toadapt object and situation models to behavioral needs not anticipatedby the human system designer. This talk gives a detailed description of  a system architecture for self-referential autonomous learning whichenables the refinement of object and situation models during operation in order to optimize behavior. This includes structural learning of hierarchical models for situations and behaviors that is triggered by a mismatch between expected and actual action outcome. Besides proposing architectural concepts the talk also describes a first implementation of the system within a simple simulated traffic scenario to demonstrate the feasibility of the approach.