In navigation tasks, mobile robots usually have to rely on onboard sensing and often have to deal with substantial uncertainty due to imperfect actuators and noisy sensor measurements. For reliable navigation, robots not only have to perform suitable movements in order to reduce their estimation uncertainty, but also need to generate safe trajectories. In this project, we develop techniques that enable efficient and safe trajectory generation by reasoning over the underlying uncertainty of the system. Therefore, we probabilistically determine safety by evaluating, for example, the risk of collision or failure that could be caused by deviations from the desired trajectory during its execution.

The main goal of the autonomous blimp project is to advance the science of autonomous microsystems through a multi-disciplinary collaboration of microsystems engineering and robotics. The blimp was developed within the PhD Program Embedded Microsystems and provides a platform to investigate the design and the application of lightweight sensors and techniques for efficient sensor data fusion under real-time conditions. Furthermore, the blimp is a challenging platform for planning and control in the context of autonomous navigation.

Modern technologies are advancing through nanoscale manufacturing technologies. Especially when minimizing supply voltage in low-power applications or in aviation and aerospace, soft errors are occurring and cause nodes within a circuit to temporarily fail. In this project, we develop techniques for efficient analysis of applications under soft errors. As a result, our approach provides means to effectively combine the robustness of target applications like probabilistic state estimation or image processing with logical hardening through error correction and redundancy for cost-saving hardware and software design.

Service robots operating in home environments or industrial settings often have to deal with dynamic and ambiguous areas. Therefore many practical mobile robot applications rely on artificial landmarks which are placed in the environment of the robot for localization. In this project, we develop algorithms for efficient landmark placement that allow the robot to operate autonomously with a target accuracy while keeping the number of landmarks needed in the environment as small as possible.

Optical motion capture systems provide accurate tracking of single markers in three-dimensional space. However, for film production or rehabilitation purposes, often knowledge about the precise configuration of the skeleton of the tracked person is required. In this project, we develop techniques for accurate human skeleton tracking which does not require one to manually label the individual markers or to train heuristic parameters for semi-automatic labeling. Furthermore, we can estimate the skeleton configuration even in the case of marker occlusions and fast movements.