Contributions in Merlin

null

null

null

We propose to use neural networks to value options when analytical solutions do not exist. The basic idea of this approach is to approximate the value function of a dynamic program by a neural net, where the selection of the network weights is done via simulated annealing. The main benefits of this method as compared to traditional approximation techniques are that there are no restrictions on the type of the underlying stochastic process and no limitations on the set of possible actions. This makes our approach especially attractive for valuing Real Options in flexible investments. We, therefore, demonstrate the method proposed by valuing flexibility for costly switch production between several products under various conditions.

In the study of adaptive behavior, autonomous mobile robots (“autonomous agents”) are meanwhile accepted as an useful research tool for gaining insight into the mechanisms underlying the behavior of natural systems. A second objective of autonomous agent research is to improve upon the state-of-the-art in engineering, in particular robotics, by incorporating design principles discovered in natural systems into the design of autonomous mobile robots. This thesis focuses on the latter objective. In autonomous agent research, the fact that animals gain much behavioral competence from dedicated peripheral structures, exploiting physical processes and analog signal processing properties, has been largely neglected so far. This is a problem because such structures are necessary for performing the kind of signal processing required for the agent to optimally interact with its environment. In this thesis it is shown that information contained in the physical structures of autonomous agents can be effectively extracted using analog signal processing techniques, allowing the agent to accomplish its given task. The method is to apply biologically inspired (neuromorphic) structures to the peripheral layer of autonomous agents, and perform real-world experiments, examining the behavioral competencies of the agents. This approach contrasts with typical robot control methods, in which information processing is performed after digitizing the peripheral signals, thereby potentially discarding information contained in the physical structure and the signal processing properties of the peripheral device. A number of case studies which investigate peripheral processing in autonomous agents are presented. The first one demonstrates that the morphology of a mobile robot can be exploited for performing a particular task (heap building) without using a dedicated control module for the task. Such kind of behavior can also be observed in certain ant species which served as inspiration source for these experiments. The second case study describes the development of polarization sensitive sensors. These sensors mimic the crossed-analyzer configuration discovered in certain insects. The outstanding advantage of this configuration is that the orientation of the perceived polarized light is measured independent of the ambient light intensity over a large dynamic range. The experimental results indicate that this neuromorphic peripheral structure can be successfully used in a celestial compass, providing directional clues for an autonomous mobile robot. The third study elaborates upon a single-chip analog sensory-motor VLSI device, which facilitates both sensor-signal processing and motor control for an autonomous mobile robot. The chip incorporates a number of neuromorphic engineering techniques, including an artificial retina and a dual feedback motor controller, allowing the agent to accomplish its linefollowing task. This implementation is based entirely on analog peripheral computing demonstrating the feasibility of robot control without the use of digital computing techniques. The fourth one discusses attention-based navigation applying controllable local sensitivity in an array of retina pixels on a silicon chip. This property is used for conditioning a particular steering direction preference for a mobile robot. In this implementation, the agent can control its peripheral device for extracting appropriate information from the retina output signal in which the visual attention-signal is encoded. The specific advantage of this realization is that there is no additional computational module necessary for obtaining the attention information. In the final study, landmark navigation using a dedicated peripheral structure implemented in analog VLSI is discussed. The objective of this development is to create a landmark navigation system based on a widely accepted model for insect navigation. In this model, the insect memorizes the scene at an important location (e.g. its nest) and uses the stored image for finding its way back to this location. With the chip is demonstrated that the algorithm describing this model can be implemented on a single peripheral device, relieving other processing modules within the agent from such computation. The experimental results demonstrate that information contained in dedicated physical peripheral structures can be exploited for achieving behavioral competencies in autonomous agents. Such mechanisms are of great importance for autonomous mobile robots because they help reduce energy consumption and improve the real-time performance of the agent. Moreover in this manner, the total computational burden of the system is distributed over a number of processing units acting in parallel, resulting in improved robustness of the system. Hence, equipping autonomous agents with structures, whose designs are based on principles discovered for peripheral signal processing in natural systems, is a very promising approach for creating intelligent behaving systems.

null

Wenn das Auftragsspektrum der Armee zunimmt und gleichzeitig die Mittel abnehmen, so zwingt dies zu einer Erhöhung der Produktivität. Gehen bei den Mitteln die Finanz- und Sachmittel zurück, so konzentriert sich die Erhöhung der Produktivität auf den Personalbereich. Soweit der Personalbereich durch das Kader geprägt wird, ist die Effizienz der Kaderausbildung zentraler Hebel für die Produktivitätserhöhung. Dies gilt für Wirtschaftsunternehmen ganz analog.