Honeybee colonies are complex social systems that consist of many individuals interacting with each other  without a central control. The organisation of tasks is determined by temporal polyethism: the task  repertoire a  bee performs changes with age. Observing individual honeybees and their interactions with each other gives insight into their collective behaviour. There are different forms of social interactions in a bee hive such as physical contact (proximity), mouth-to-mouth food exchange (trophallaxis) and waggle dancing. An approach to the analysis on interactions is constructing a social network, where bees form the nodes and interactions the edges of the network. Other works have already analysed netowrks deriving from interactions in a honeybee hive. So far, no work has compared the trophallaxis and proximity network measures using data of all bees in the colony.

The automatically collected data of a honeybee colony by the BeesBook Project is used to construct the network. The biological age and network age, capturing a bee's current role in the hive, of each bee is available.

The goal of this thesis is to analyse and quantify the differences of the proximity network and the trophallaxis network with a focus on task repertoire. One part of the analysis is exploring the daily interactions between pairs of bees using date of one week. Afterwards, network measures are calculated for the proximity and trophallaxis network based on the data of one day. Finally, networks for every four hours of the day are compared.

While the analysis of the weekly data shows a high similarity between both types of interactions, the analysis of network measures reveals great dissimilarities. For the trophallaxis network, the network measures of nodes are negatively correlated to network age. In the proximity network, on the other hand, young bees have the highest values, but older bees also show high values. The proximity interactions of old bees are dependent on the circadian cycle, whereas the trophallaxis interactions do not vary strongly at different times of the day.

The results for both networks suggest a strong link between network measures and taks allocation in a honeybee colony.

For autonomous vehicles it is important to be able to detect traffic lights. This work suggests a method for the Freie Universität Berlin's AutoMiny model cars to detect traffc lights. This method uses image thresholding to find candidate regions and uses the normalized rgb colors of the candidate region to check if a traffic light was found. The evaluation shows that this method is able to reliably detect traffic lights in short distance but struggles with traffic lights further away. Also a detection in real-time is not possible with the hardware of the car.

Autonomous driving has become very popular over the years. To successfully develop a car that drives by itself and more importantly follows the traffic laws, many necessary factors need to be considered. One important factor ist he detection of traffic signs in order to be able to react properly. Thus, one component of self-driving cars are traffic sign detection systems.

Such a detection system has been developed for the model cars of the research group autonomous cars using support vector machines and the histogram of oriented gradients. For training and evaluating the systems, a new dataset consisting of images, made in the robotic laboratory, has been created. In addition to the dataset, recorded rosbag files of a driving model car has been used for the evaluation.

The developed system is able to classify six traffic signs and reaches a high precision. The recall, however, needs to be improved. In order to achieve an improvement, a few adaptations have been suggested at the end.

Self-driving cars are the next step towards safe and convenient travel, but, as with all machine learning applications, require loads of training data. It would be desireable if the Freie Universität Berlin could us readily available datasets to prototype new machine learning models instead of creating their own dataset from test drives of one of their self-driving-cars.

Multiple traffic light detection models were trained on the popular datasets BSTLD and DTLD using the Tensorflow Research repository. The evaluation revealed that predictive power achieved in one dataset generally transfers over to another similar dataset with minimal loss in performance. Neither geographical differences between datasets (e.g. traffic lights at the beginning or the end of an intersection) nor architecture choices seem to impact this result. Some ideas to further reduce performance penalties are given for future work.