Timo Korthals

A hierarchical system for word discovery exploiting DTW-based initialization

Discovering the linguistic structure of a language solely from spoken input asks for two steps: phonetic and lexical discovery. The first is concerned with identifying the categorical subword unit inventory and relating it to the underlying acoustics, while the second aims at discovering words as repeated patterns of subword units. The hierarchical approach presented here accounts for classification errors in the first stage by modelling the pronunciation of a word in terms of subword units probabilistically: a hidden Markov model with discrete emission probabilities, emitting the observed subword unit sequences. We describe how the system can be learned in a completely unsupervised fashion from spoken input. To improve the initialization of the training of the word pronunciations, the output of a dynamic time warping based acoustic pattern discovery system is used, as it is able to discover similar temporal sequences in the input data. This improved initialization, using only weak supervision, has led to a 40% reduction in word error rate on a digit recognition task.

Semantical occupancy grid mapping framework

In recent decades, mapping has been increasingly investigated and applied in unmanned terrain, aerial, sea, and underwater vehicles. While exploiting various mapping techniques to build an inner representation of the environment, one of the most famous remaining is occupancy grid mapping. It has been applied to all domains in a 2D/3D fashion for localization, mapping, navigation, and safe path traversal. Until now generally active range measuring sensors like LiDAR or SONAR are exploited to build those maps. With this work the authors want to overcome these barriers by presenting an occupancy mapping framework offering a generic sensor interface. The interface handles occupancy grids as inverse sensor models, which may represent knowledge on different semantical decision levels and therefore build up a semantic grid map stack. The framework offers buffered memory management for efficient storing and shifting and further services for accessing the 2D map stack in different cell-wise pre-fused and topometric ways. Within the framework, two novel techniques operating especially with occupancy grids are presented: First, a novel odds based interpolation filter is introduced, which scales grid maps in a Bayesian way. Second, a Supercell Extracted via Variance-Driven Sampling (SEVDS) algorithm is presented which, abstracts the semantical occupancy grid stack to a topometric map. While this work focuses on the frameworks introduction, it is extended by the evaluation of SEVDS against state-of-the-art superpixel approaches to prove its applicability.

ToBI – Team of Bielefeld: Enhancing Robot Behaviors and the Role of Multi-robotics in RoboCup@Home

In this paper, we describe the joint effort of the Team of Bielefeld (ToBI) which won the RoboCup@Home competition in Leipzig 2016. RoboCup@Home consists of a defined set of benchmarking tests that cover multiple skills needed by service robots. We present the robotic platforms, technical contributions, and lessons learned from previous events that led to the final success this year. This includes a framework for behavior modeling and communication employed on two human-sized robots Floka and Biron as well as on the small robotic device AMiRo. These were used for a multi-robot collaboration scenario in the Finals. We describe our main contributions in automated testing, error handling, memorization and reporting, robot-robot coordination, and flexible grasping that considers object shape.

AMiRo: A modular & customizable open-source mini robot platform

AMiRo is a novel modular robot platform that can be easily extended and customized in hardware and software. Built up of electronic modules that fully exploit recent technology and open-source software for sensor processing, actuator control, and cognitive processing, the robot facilitates rich autonomous behaviors. Further contribution lies in the completely open-source software habitat: from low-level microcontroller implementations, over high-level applications running on an embedded processor, up to hardware accelerated algorithms using programmable logic. This paper describes in detail the motivation, system architecture, and software design of the AMiRo, which surpasses state-of-the-art competitors.

Evidence Grid Based Information Fusion for Semantic Classifiers in Dynamic Sensor Networks

We propose an anytime fusion setup of anonymous distributed information sources with spatial affiliation. For this approach we use the evidence grid mapping algorithm which allows to fuse sensor information by their inverse sensor model. Furthermore, we apply an online Mixture of Experts training such that faulty voters are detected and suppressed during runtime by a gating function.

AMiRo: A Mini Robot for Scientific Applications

The Autonomous Mini Robot (AMiRo) is a modular and extensible mini robot platform, designed for scientific research and education. Its decentralized architecture enables to easily add or remove functionalities as required for any application. A well defined physical and electrical interface offers the possibility to design new modules with minimal effort. The open-source software framework for the AMiRo is already growing, since the robot is commonly used for research, education, and competitions. Several demonstrations of the system are given, which present its capabilities. Starting with a fuzzy controller for line following, these demonstrations include remote controlling as well as an implementation of an artificial neural network running on the platform.

AMiRo: A Mini Robot as Versatile Teaching Platform

Since robots become increasingly ubiquitous and system complexity increases, teaching university students in robotics is essential for modern studies in computer science. This work thus presents the education curriculum around the Autonomous Mini Robot (AMiRo) as a solution to this challenge. The goal is to provide insights to the various fields related to robotics and allow students to specialize in a wide range of topics, depending on their interests. Concept as well as platform have been evaluated and the results reveal a generally positive feedback as well as some issues, for which according solutions are proposed.

Generic Architecture for Modular Real-time Systems in Robotics

With the continuous progress in robotics and application of such systems in evermore scenarios, safety and flexibility become increasingly important aspects and new designs should thus emphasize real-time capability and modularity. This work points out all related topics for such an endeavor and proclaims to move from conventional bottom-up design to more holistic approaches. Based on experience gained with the modular mini robot platforms BeBot and AMiRo, a novel generic modular architecture is proposed that offers high flexibility and system wide real-time capability.

Multi-Modal Detection and Mapping of Static and Dynamic Obstacles in Agriculture for Process Evaluation

Today, agricultural vehicles are available that can automatically perform tasks such as weed detection and spraying, mowing, and sowing while being steered automatically. However, for such systems to be fully autonomous and self-driven, not only their specific agricultural tasks must be automated. An accurate and robust perception system automatically detecting and avoiding all obstacles must also be realized to ensure safety of humans, animals, and other surroundings. In this paper, we present a multi-modal obstacle and environment detection and recognition approach for process evaluation in agricultural fields. The proposed pipeline detects and maps static and dynamic obstacles globally, while providing process-relevant information along the traversed trajectory. Detection algorithms are introduced for a variety of sensor technologies, including range sensors (lidar and radar) and cameras (stereo and thermal). Detection information is mapped globally into semantical occupancy grid maps and fused across all sensors with late fusion, resulting in accurate traversability assessment and semantical mapping of process-relevant categories (e.g., crop, ground, and obstacles). Finally, a decoding step uses a Hidden Markov model to extract relevant process-specific parameters along the trajectory of the vehicle, thus informing a potential control system of unexpected structures in the planned path. The method is evaluated on a public dataset for multi-modal obstacle detection in agricultural fields. Results show that a combination of multiple sensor modalities increases detection performance and that different fusion strategies must be applied between algorithms detecting similar and dissimilar classes.