|Beschreibung:||The aim of this thesis is to develop a system that enables autonomous and situated agents to learn and adapt to the environment in which they live and operate. In doing so, the system exploits both adaptation through learning and evolution. A unified approach to learning and adaptation, which combines the principles of neural networks, reinforcement learning and evolutionary methods, is used as a basis for the development of the system. In this regard, a novel method, called Evolutionary Acquisition of Neural Topologies (EANT), of evolving the structures and weights of neural networks is developed. The method introduces an efficient and compact genetic encoding of a neural network onto a linear genome that encodes the topology of the neural network implicitly in the ordering of the elements of the linear genome. Moreover, it enables one to evaluate the neural network without decoding it. The presented genetic encoding is complete in that it can represent any type of neural network. In addition to this, it is closed under both structural mutation and a specially designed crossover operator which exploits the fact that structures originating from some initial structure have some common parts. For evolving the structure and weights of neural networks, the method uses a biologically inspired meta-level evolutionary process where new structures are explored at larger timescale and existing structures are exploited at smaller timescale. The evolutionary process starts with networks of minimal structures whose initial complexity is specified by the domain expert. The introduction of neural structures by structural mutation results in a gradual increase in the complexity of the neural networks along the evolution. The evolutionary process stops searching for the solution when a solution with the necessary minimum complexity is found. This enables EANT to find optimal neural structures for solving a given learning task. The efficiency of EANT is tested on couple of learning tasks and its performance is found to be very good in comparison to other systems tested on the same tasks.