Tobias Schubert
Research
Multi Robot Localization
Localization is one of the most fundamental tasks for mobile robots. For a team of multiple robots, a paradigm called Collaborative Localization (CL) has been demonstrated to provide significant improvements of the localization performance of the individual teammates. In CL, robots detect each other and communicate their estimates, which correlates the estimates of their individual poses. It is of fundamental importance to keep track of these dependencies to avoid the problem of double-counting or data-incest, which occurs when two robots treat shared information as uncorrelated.
In many relevant multi-robot applications -- for example underwater, in mines, or in large-scale environments, -- communication might be energy consuming, error-prone, slow, or simply not possible at all times. Therefore, decentralized architectures that reduce the need for communication to a minimum are desirable.
We restrict the communications to places where the robots actually meet. Thus, exchange of information takes place only between two
robots obtaining a relative measurement. We developed an EKF-based localization algorithm that is able to approximate the covariance
under the above-mentioned constraint.
MAKE IT ReaL - Motion capture tracking of humans and animals
The aim in this project is the development of an online motion capture tracking system, which is able to track complicated human and animal motions in a robust way. More precisely we glue indistinguishable markers on arbitrary positions of the skin of the mammal and measure their positions in motion. The main task are to initialize the method, to associate the resulting data positions to the right markers and find the underlying skeleton structure.
While many existing approaches can deal with the latter two problems, they typically need a specific pose for initialization. As this is rather unpractical in the context of animal tracking this often requires a manual initialization process. We developed an approach to reliably track animals and humans in marker-based optical motion capture systems with freely attached markers that is also able to perform an automatic initialization without any pre- or post-processing of the data.Many existing skeleton tracking approaches use a predefined skeleton structure, which is globally scaled to the persons or animals height. Detailed skeleton parameter studies only exists for some kind of animals and additionally there is a lot of variation in the skeleton bone parameter relations between different mammal breeds. We developed an approach for an automatic bone parameter estimation during tracking.
NaRKo - Nachgiebige Serviceroboter für Krankenhauslogistik (Compliant Service Robots for Logistics in Hospitals)
Service robots which act autonomously in an indoor environment are faced with challenging problems. First of all they need to navigate in crowded environments in social compliant way. Due to small hallways in common hospitals collision free paths can not always be found by classic motion planning algorithms. Even recently invented motion planning algorithms, which incorporate dynamic obstacles as cooperative agents, lead to frozen robot scenarios. This implies that one can not assign tasks to the robot which have a time constraint, which questions the whole service robotics in crowded environments.
In this project we challenge the problem of compliant mobile service robots. Humans typically use the speech and apply small contact forces with their hands to navigate through crowded environments. Our goal is to adopt the later concept, namely to allow human-robot collisions and to incorporate contact forces in the motion planning. Thereby, we use a whole body sensory concept to record the forces. Human use contact forces rarely for navigation, they use highly developed situation analysis techniques to avoid collision. We will use visual perception for detection and tracking of humans and dynamic objects and we will improve existing methods for situation analysis. Fully compliant mobile robotic systems require omni-directional drive. Besides the mobile robotic system should be able to carry heavy weight load like the dishes or hospital beds. In this project we will invent a low cost omni-directional platform, which is scalable and robust.
DataGlove - Detection of fine motor disabilities of stroke patients
Direct after a stroke many patients have motor disabilities. These disabilities constrain their normal live for a long time after the stroke. So a fast regain of the former abilities is a desirable issue. In order to evaluate medical treatments an objective measure is needed. In this project we use a group of stroke patients, which have an unilateral disability of one hand. These stroke patients have to execute motor tests, like the nine hole peg test, while wearing data gloves from CyberGlove. Then we compare the differences in movement strategies, trajectories and acceleration of the healthy hand to the unhealthy hand. Thereby we use a Gaussian process dynamical model.
CaliMotion - Motion analysis for Parkinson patients
The objective characterization of human motion is required in a variety of fields including competitive sports, rehabilitation and the detection of motor deficits. In this project we are faced with the motion analysis of Parkinson patients. Deep brain stimulation is widely used as a treatment for advanced Parkinson's disease, but requires stimulation parameter adjustment from time to time. The goal of this project is to automatically adjust the stimulation parameter based on motion capture data. Nowadays, typically human experts evaluate the motor behavior. These evaluations are based on their individual experience which leads to a low inter- and intra-expert reliability. Standardized tests improve on the reliability but are still prone to subjective ratings and require human expert knowledge. We develop a method to characterize the motor state of Parkinson patients using full body motion capturing data. Our approach merges various metrics with a Random Forest and uses a probabilistic formulation to compute a one-dimensional measure for the performed motion. Overall we obtained an objective performance measure competitive with the Unified Parkinson’s Disease Rating Scale (UPDRS).