Almost all of today’s knowledge is stored in databases and thus can only be accessed with the help of domain specific query languages, strongly limiting the number of people which can access the data. In this work, we demonstrate an end-to-end trainable question answering (QA) system that allows a user to query an external NoSQL database by using natural language. A major challenge of such a system is the non-differentiability of database operations which we overcome by applying policy-based reinforcement learning. We evaluate our approach on Facebook’s bAbI Movie Dialog dataset and achieve a competitive score of 84.2% compared to several benchmark models. We conclude that our approach excels with regard to real-world scenarios where knowledge resides in external databases and intermediate labels are too costly to gather for non-end-to-end trainable QA systems.

This paper focuses on the comparison of the low-level sensor failure detection capabilities of model-based Kalman filtering and a model-free Fuzzy voting approach. The ability to identify failing and degrading sensors is essential when dealing with error-prone data acquired in harsh application environments as can be typically found in the field of embedded systems. In order to investigate the respective performance concerning rejection of faulty data several experiments were conducted in a simulation environment. The two candidates were selected to gain more insight into the advantages and limitations of system modeling (or the lack thereof) in signal-level data fusion. The Kalman filter was selected as a typical candidate that relies on extensive models for both the system and information sources. The fuzzy approach, however, employs a heuristic that requires no modeling at all. This results in a broader field of possible applications since detailed knowledge is no longer required. Thus, it can be employed in scenarios that one would not be able to use a model-based algorithm. Such applications include scenarios with ongoing reconfiguration (e.g. wireless sensor networks) or systems with limited detail knowledge about the devices.

In this paper the compliant low level control of a biologically inspired control architecture suited for bipedal dynamic walking robots is presented. It consists of elastic mechanics, a low-level compliant joint controller and a hierarchical reflex-based control layer. The former is implemented on a DSP while the reflex network is located on a desktop PC. Thus, one is able to utilize distribution as a powerful means to guarantee low latency and scalability. The concept is tested on a prototype leg mounted on a vertical slider that is designed to perform cyclic squat jumps. Thus, a suited mechatronic setup that features highly dynamic actuators as well as energy storage capabilities is derived. Cyclical jumping is employed as a benchmark for the system's performance. Experimental results of the prototype setup as well as simulation runs are presented and compared to human squat jumping.

This paper introduces our recently developed high precision robot localization system employing position sensitive detectors (PSD). A lateral effect PSD is an ideal position sensitive device with a resolution of approx. 0.5 mum. Such a device can be used to enhance the precision capabilities of robot localization or robot calibration. The developed PSD guided localization system is comprised of a structured laser pointer system attached to the robot end-effector and a PSD fixture. It is capable of feeding back the movement information of the TCP relative to the PSD frame. This is achieved by detecting the position of the laser beams emitted from the laser pointers onto the PSD surface. The system was well designed, mapped, and constructed and hence allows it to perform precision localization of the robot. Simulations and preliminary experiments demonstrated the effectiveness of the developed PSD guided localization system including hardware, software, and algorithms. Therefore, it is capable of improving the localization performance of robots as well as promoting the application of precision robot calibration.