This dissertation contributes to the field of biomimetic robotics , which focuses on robotics applications inspired by biology. The imitation of biological structure or function has often proven useful in solving problems that cannot be solved with classical robotics techniques. The dissertation provides a novel example of biological mimicry for robot navigation, and describes a unified framework within which many biomimetic robotics applications can be studied and compared.;The first part of the dissertation proposes a method for reactive mobile robot navigation in unstructured and non-stationary environments. A neural network learns to control avoidance and approach behaviors in a mobile robot, closely imitating two forms of animal learning known as classical conditioning and operant conditioning. The use of animal learning and other biomimetic techniques for robot control leads to several desirable properties that distinguish this approach from other robotics approaches. First, the robot learns without supervision to identify informative cues in its environment and to predict the consequences of its own actions. Second, learning is fast and occurs simultaneously for these opposite behaviors. Third, after learning the robot can navigate robustly through unknown or constantly changing environments while avoiding obstacles and approaching sources of light. Fourth, the neural network requires no knowledge of the geometry of the robot or of the quality and configuration of the robot's sensors. In other words, the model is platform-independent, and it can be used with a variety of robotics platforms essentially with no modifications, as demonstrated in this dissertation.;Many examples of biomimetic robotics applications have been proposed in the last five decades. However, until recently there has been no coordinated attempt to summarize all these applications in a unified framework. The second part of this dissertation attempts to define the field of biomimetic robotics by reviewing a large number of applications and classifying them within a taxonomy similar to the taxonomy used by biologists to classify the animal kingdom. The functional and structural similarities between different applications are also summarized. Finally, the model described in the first part of the dissertation is discussed in the context of the proposed framework.
展开▼
