direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Robotics Related

Environment Design in Motion Planning

13. April 2018

Bild

Exploiting the environment is useful to reduce robot state uncertainty due to very reliable force feedback. Therefore, we extended an RRT based motion planner with motion actions which allow the planner to make contact with the environment. This raised the importance of actual shape and availability of environment features to enable contacts. A first environment design algorithm has been developed that iteratively adds new environment features to a given scene. To do so, the algorithm detects and models areas of high uncertainty to then constructs and places landmarks into the uncertain areas. more to: Environment Design in Motion Planning

Life-long grasp success learning

25. July 2018

Bild

Grasping is a hard task for robots, because it is impossible to model the world and the robot dynamics perfectly. Without exhaustive modeling, we can increase the grasp performance of the robot by exploiting features of the environment like the wall of a crate. This features are called environment constraints (EC). The robot can push the object to one of the walls, by which the object motion is constrained along the wall and its location uncertainty is also reduced. more to: Life-long grasp success learning

Predicting Interaction Points for Robot Exploration from RGB-D Data

14. August 2017

Bild

A large part of our everyday environment consist of rigid bodies and kinematic joints that connect them. Together these joints and bodies form kinematic structures, such as cupboards, drawers, doors etc. and robots that work in our homes or offices need to be able to explore and analyze such kinematic structures. In order to reveal and learn about kinematic joints, robots first need to make the objects attached to these joints move. E.g. in order to estimate the direction in which a door opens, a robot needs to first grasp the handle of that door and then pull or push it. Only as soon as it starts to move, the robot can estimate where the rotational axis of that door is. Before the robot can actually grasp and excert forces onto an object, it needs to create hypotheses where and how to grasp. The focus of this thesis would be on generating such hypotheses. more to: Predicting Interaction Points for Robot Exploration from RGB-D Data

Predicting Kinematic Degrees of Freedom from RGB-D Data

14. August 2017

Bild

A large part of our everyday environment consist of rigid bodies and kinematic joints that connect them. Together these joints and bodies form kinematic structures, such as cupboards, drawers, doors etc. and robots that work in our homes or offices need to be able to explore and analyze such kinematic structures. An important aspect of this exploration is the generation of initial hypotheses about the location and parameters of kinematic joints, based on visual input. If a robot can generate a good set of such initial hypotheses, this can guide its exploration and limit the number of actions that it needs to reveal kinematic degrees of freedom. In this thesis you develop a method for predicting kinematic degrees of freedom from RGB-D data. more to: Predicting Kinematic Degrees of Freedom from RGB-D Data

Motion Planning Under Uncertainty for Soft Manipulation

20. April 2017

Bild

Uncertainty is the major obstacle for robots manipulating objects in the real world. A robot can never perfectly know its position in the world, the position of objects, and the outcome of its actions. A particularly hard challenge is motion planning under uncertainty. How should the robot move, if the model of the world might be wrong or incomplete? However, a robot can significantly reduce uncertainty if it uses contact sensing to establish controlled contact with the environment. Imagine a robot pushing objects into an edge of the environment - this action will reduce uncertainty over all objects positions. more to: Motion Planning Under Uncertainty for Soft Manipulation

Computational Biology Related

Leveraging Data from Mass Spectrometry to Improve Cross-Links

Bild

Cross-Links are experimentally derived distance constraints that can be used to guide protein structure prediction. The current cross-linking data is obtained from multiple mass spectrometer runs and pre-filtered. In the pre-fltering step we lose a lot of information. For instance, there is ambiguity in the data that is currently handled in a greedy fashion. The idea of this project is to go one step back and look at the "raw" data from the mass spectrometry experiments. The hope is by including external information and dealing directly with the ambiguity in the data we can improve the number of cross-links and/or need fewer experiments for a similar yield. more to: Leveraging Data from Mass Spectrometry to Improve Cross-Links

Leveraging Crosslinks for Template Retrieval

Bild

Cross-Link Mass Spectometry is an experimental technique that provides residue distance constraints on the native structure. The method has been previously used to guide ab initio protein structure prediction. You will develop in this thesis a method that leverages cross-links to find homologous structures to the target in the PDB. For that, you will compare the distance constraints provided by cross-links with simulated cross-links for the templates from the PDB. more to: Leveraging Crosslinks for Template Retrieval

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe