Sven Hellbach

Task-level imitation learning using variance-based movement optimization

Recent advances in the field of humanoid robotics increase the complexity of the tasks that such robots can perform. This makes it increasingly difficult and inconvenient to program these tasks manually. Furthermore, humanoid robots, in contrast to industrial robots, should in the distant future behave within a social environment. Therefore, it must be possible to extend the robots abilities in an easy and natural way. To address these requirements, this work investigates the topic of imitation learning of motor skills. The focus lies on providing a humanoid robot with the ability to learn new bi-manual tasks through the observation of object trajectories. For this, an imitation learning framework is presented, which allows the robot to learn the important elements of an observed movement task by application of probabilistic encoding with Gaussian Mixture Models. The learned information is used to initialize an attractor-based movement generation algorithm that optimizes the reproduced movement towards the fulfillment of additional criteria, such as collision avoidance. Experiments performed with the humanoid robot ASIMO show that the proposed system is suitable for transferring information from a human demonstrator to the robot. These results provide a good starting point for more complex and interactive learning tasks.

Large scale place recognition in 2D LIDAR scans using Geometrical Landmark Relations

The recognition of places that have already been visited is a fundamental requirement for a mobile robot. This particularly concerns the detection of loop closures while mapping environments as well as the global localization w.r.t. to a prior map. This paper introduces a novel solution to place recognition with 2D LIDAR scans. Existing approaches utilize descriptors covering the local appearance of discriminative features within a bag-of-words (BOW) framework accompanied with approximate geometric verification. Though limiting the set of potential matches their performance crucially drops for increasing number of scans making them less appropriate for large scale environments. We present Geometrical Landmark Relations (GLARE), which transform 2D laser scans into pose invariant histogram representations. Potential matches are found in sub-linear time using an efficient Approximate Nearest Neighbour (ANN) search. Experimental results obtained from publicly available datasets demonstrate that GLARE significantly outperforms state-of-the-art approaches in place recognition for large scale outdoor environments, while achieving similar results for indoor settings. Our Approach achieves recognition rates of 93% recall at 99% precision for a dataset covering a total path of about 6.5 km.

Sparse coding of human motion trajectories with non-negative matrix factorization

We use shift-invariant Non-negative Matrix Factorization (NMF) for decomposing continuous-valued time series into a number of characteristic primitives, i.e. the basis vectors, and their activations, which results in a model-independent and fully data driven parts-based representation. We interpret the basis vectors as short parts of motion that are shared between all trajectories in the data set, and the activations as onset times of those parts. The extension of the shift-invariant NMF by a new competition term between adjacent activations allows to gain temporally isolated activation events, which further supports this interpretation. We show that the resulting sparse and compact representation can be used for the prediction of motion trajectories, and that it can be beneficial for classification, because it allows the application of simple standard classification models with few parameters. In this paper we show that basis vectors can be extracted, which can be interpreted as short motion segments. We present results on trajectory prediction, and show that the sparse representation can be used for classification of trajectories of a single joint, like the one of a hand, obtained by motion capturing.

Whom to talk to? Estimating user interest from movement trajectories

Correctly identifying people who are interested in an interaction with a mobile robot is an essential task for a smart Human-Robot Interaction. In this paper an approach is presented for selecting suitable trajectory features in a task specific manner from a huge amount of different forms of possible representations. Different sub-sampling techniques are proposed to generate trajectory sequences from which features are extracted. The trajectory data was generated in real world experiments that include extensive user interviews to acquire information about user behaviors and intentions. Using those feature vectors in a classification method enables the robot to estimate the users interaction interest. For generating low-dimensional feature vectors, a common method, the Principle Component Analysis, is applied. The selection and combination of useful features out of a set of possible features is carried out by an information theoretic approach based on the Mutual Information and Joint Mutual Information with respect to the users interaction interest. The introduced procedure is evaluated with neural classifiers, which are trained with the extracted features of the trajectories and the user behavior gained by observation as well as user interviewing. The results achieved indicate that an estimation of the users interaction interest using trajectory information is feasible.

Feel like an insect: a bio-inspired tactile sensor system

Insects use their antennae (feelers) as near range sensors for orientation, object localization and communication. This paper presents an approach for an active tactile sensor system. This includes a new type of hardware construction as well as a software implementation for interpreting the sensor readings. The discussed tactile sensor is able to detect an obstacle and its location in 3D space. Furthermore the material properties of the obstacles are classified by use of neural networks.

Echo State Networks for Online Prediction of Movement Data --- Comparing Investigations

This papers intention is to adapt Echo State Networks to problems being faced in the field of Human-Robot Interactions. The idea is to predict movement data of persons moving in the local surroundings by understanding it as time series. The prediction is done using a black box model, which means that no further information is used than the past of the trajectory itself. This means the suggested approaches are able to adapt to different situations. For experiments, real movement data as well as synthetical trajectories (sine and Lorenz-attractor) are used. Echo State Networks are compared to other state-of-the-art time series analysis algorithms, such as Local Modeling, Cluster Weighted Modeling, Echo State Networks, and Autoregressive Models. Since mobile robots highly depend on real-time application.

Awakening history: Preparing a museum tour guide robot for augmenting exhibits

While the idea of tour guide robots has been addressed several times, applying a video-projector based augmented reality component to this scenario is rather new. We show requirements of the localization system of the robot and how they can be fulfilled, as well as a basic system for projection correction and its suitability for this scenario.

Wizard of Oz revisited: Researching on a tour guide robot while being faced with the public

In this paper, we propose the usage of a Wizard of Oz to facilitate the first steps towards a tour-guide robot in a real world environment. In particular, we want to emphasize two essential aspects during such an early project phase. On the one hand, to ensure certain basic skills of the robot (e.g. navigation and interaction), suitable methods have to be selected. On the other hand, insights about the requirements of further developments have to be accumulated. At the same time, it should be avoided to annoy the visitors of the museum by malfunctioning systems. Hence, the robot has to appear to be performing flawlessly. Thus, in such early project stages, we suggest to substitute the still missing parts of the robot by a human operator.

Tactile conditioning and movement analysis of antennal sampling strategies in honey bees (Apis mellifera L.).

Honey bees (Apis mellifera L.) are eusocial insects and well known for their complex division of labor and associative learning capability(1, 2). The worker bees spend the first half of their life inside the dark hive, where they are nursing the larvae or building the regular hexagonal combs for food (e.g. pollen or nectar) and brood(3). The antennae are extraordinary multisensory feelers and play a pivotal role in various tactile mediated tasks(4), including hive building(5) and pattern recognition(6). Later in life, each single bee leaves the hive to forage for food. Then a bee has to learn to discriminate profitable food sources, memorize their location, and communicate it to its nest mates(7). Bees use different floral signals like colors or odors(7, 8), but also tactile cues from the petal surface(9) to form multisensory memories of the food source. Under laboratory conditions, bees can be trained in an appetitive learning paradigm to discriminate tactile object features, such as edges or grooves with their antennae(10, 11, 12, 13). This learning paradigm is closely related to the classical olfactory conditioning of the proboscis extension response (PER) in harnessed bees(14). The advantage of the tactile learning paradigm in the laboratory is the possibility of combining behavioral experiments on learning with various physiological measurements, including the analysis of the antennal movement pattern.

What's around me: Towards non-negative matrix factorization based localization

This paper presents a new method for representing the local and global environment captured by 2D range scans using non-negative matrix factorization (NMF). Unlike other approaches, we do not predefine features or geometric primitives, but, in contrast, extract environment specific basis primitives from occupancy grid maps. Similar to feature based methods, these enable fast map matching and simultaneously allow to reconstruct the entire initial map. Our approach enables efficient localization on highly compact environment representations. The proposed approach is applied to a publicly available dataset. First promising results are presented motivating further investigation in the application of NMF for localization and mapping.