Bilal Wehbe

ROBOCADEMY — A European Initial Training Network for underwater robotics

In 2014, funded by the European Commission through the Marie Curie Programme, ten leading European research institutes and companies in underwater robotics formed the ROBOCADEMY Initial Training Network (ITN). The objective of the network is to educate young researchers from Europe and abroad in the development and application of underwater robots. The ROBOCADEMY training programme comprises of scientific as well as soft-skills courses. Hands-on training is provided through integration in interdisciplinary project-teams and secondments to industry. In their PhD research projects, the ROBOCADEMY fellows develop key enabling technologies for the scientific action lines of disturbance rejection, preception and autonomy. The overall scientific goal of the project is to contribute to the next generation of resilient and robust Autonomous Underwater Vehicles (AUVs). This paper provides a brief introduction and overview in the concept of the ROBOCADEMY training network and the scientific resarch topics addressed.

Experimental evaluation of various machine learning regression methods for model identification of autonomous underwater vehicles

In this work we investigate the identification of a motion model for an autonomous underwater vehicle by applying different machine learning (ML) regression methods. By using the data collected from the robots on-board navigation sensors, we train the regression models to learn the damping term which is regarded as one of the most uncertain components of the motion model. Four regression techniques are investigated namely, artificial neural networks, support vector machines, kernel ridge regression, and Gaussian processes regression. The performance of the identified models is tested through real experimental scenarios performed with the AUV Leng. The novelty of this work is the identification of an underwater vehicles motion model, for the first time, through machine learning methods by using the robots onboard sensory data. Results show that the damping model learned with nonlinear methods yield better estimates than the simplified linear and quadratic model which is identified with least-squares technique.

AUV x — A novel miniaturized autonomous underwater vehicle

Wiihìn the 3 year project DAEDALUS the battery powered underwater vehicle AUVx was developed at DFKI. This autonomous underwater vehicle (AUV) is a novel, miniaturized exploration and research vehicle (see Fig. 1). The AUVx can be operated both autonomous or remotely as a hybrid ROV with a near field optical communication modem or a copper wire cable. What makes this paper unique is that it describes the development of a system which has a high degree of miniaturization and the feature richness of much bigger systems: A complex heterogeneous processing architecture, sensor fusion of inertial data, a mathematical model of the robot and visual data which improves the AUVs pose and orientation estimation, a unique propulsion concept and the possibility to hover statically in the water column. Results of the design and build process with a strong focus on the CFD analysis and the evaluated model are presented. The system was successfully tested in water basins at DFKI.

Learning coupled dynamic models of underwater vehicles using Support Vector Regression

This work addresses a data driven approach which employs a machine learning technique known as Support Vector Regression (SVR), to identify the coupled dynamical model of an autonomous underwater vehicle. To train the regressor, we use a dataset collected from the robots on-board navigation sensors and actuators. To achieve a better fit to the experimental data, a variant of a radial-basis-function kernel is used in combination with the SVR which accounts for the different complexities of each of the contributing input features of the model. We compare our method to other explicit hydrodynamic damping models that were identified using the total least squares method and with less complex SVR methods. To analyze the transferability, we clearly separate training and testing data obtained in real-world experiments. Our presented method shows much better results especially compared to classical approaches.