Cognitive Maps for Autonomous Machines in Construction and Civil Engineering

This thesis addresses the question of how the development process of software systems for the task of scene perception and interpretation in weakly-structured outdoor applications can be improved. It presents a novel approach for the definition and implementation of systems for perception and cognition tasks tailored especially to the difficult conditions in the domain of Robotics for Construction and Civil Engineering. In contrast to other approaches, the developed Cognitive Map Architecture avoids the notion of objects and proposes a processing scheme based on human and animal cognition, using parallel data flow networks with a strong emphasis on abstraction and unification of data to set up structures in accordance with constructionist models of cognition. The base principle of the proposed approach is that the full process of cognition consists of separable, highly parallel stages. The derivation of the concept is supported by examples of experiments which explain different mechanisms of human cognition. The proposed approach is demonstrated in various application examples in the context of an autonomous construction vehicle, a full-scale bucket excavator. The architecture is universally extensible and suitable for other applications which require the extraction, combination, and comparison of multi-modal data.