A5022626150- Muscle activation and electromyography studies
- Robot Manipulation and Learning
- Neural Networks and Applications
- EEG and Brain-Computer Interfaces
- Gaussian Processes and Bayesian Inference
- Motor Control and Adaptation
- Advanced Vision and Imaging
- Domain Adaptation and Few-Shot Learning
- Generative Adversarial Networks and Image Synthesis
- Reinforcement Learning in Robotics
- Human Pose and Action Recognition
- Model Reduction and Neural Networks
- Neuroscience and Neural Engineering
- Gamma-ray bursts and supernovae
- Robotics and Sensor-Based Localization
- Hand Gesture Recognition Systems
- Advanced Image and Video Retrieval Techniques
- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Robotic Mechanisms and Dynamics
- Fault Detection and Control Systems
- Prosthetics and Rehabilitation Robotics
- Machine Learning and Data Classification
- Vestibular and auditory disorders
- Machine Learning and Algorithms
Volkswagen Group (Germany)
2017-2024
Eötvös Loránd University
2020-2024
Ludwig-Maximilians-Universität München
2020-2022
Volkswagen Group (United States)
2017-2021
University of Tübingen
2018
Data:Lab Munich (Germany)
2017-2018
BioMimetic Systems (United States)
2018
Technical University of Munich
2012-2017
Fortiss
2015-2017
Universidad de Granada
2016
Convolutional neural networks (CNNs) have recently been very successful in a variety of computer vision tasks, especially on those linked to recognition. Optical flow estimation has not among the tasks CNNs succeeded at. In this paper we construct which are capable solving optical problem as supervised learning task. We propose and compare two architectures: generic architecture another one including layer that correlates feature vectors at different image locations. Since existing ground...
Abstract With the increased focus on quantum circuit learning for near-term applications devices, in conjunction with unique challenges presented by cost function landscapes of parametrized circuits, strategies effective training are becoming increasingly important. In order to ameliorate some these challenges, we investigate a layerwise strategy circuits. The depth is incrementally grown during optimization, and only subsets parameters updated each step. We show that when considering...
This paper is about (self-powered) advanced hand prosthetics and their control via surface electromyography (sEMG). We hereby introduce to the biorobotics community first version of Ninapro database, containing kinematic sEMG data from upper limbs 27 intact subjects while performing 52 finger, wrist movements interest. The setup experimental protocol are distilled existing literature thoroughly described; then analysed results discussed. In particular, it clear that standard analysis...
Convolutional neural networks have been applied to a wide variety of computer vision tasks. Recent advances in semantic segmentation enabled their application medical image segmentation. While most CNNs use two-dimensional kernels, recent CNN-based publications on featured three-dimensional allowing full access the structure images. Though closely related segmentation, includes specific challenges that need be addressed, such as scarcity labelled data, high class imbalance found ground truth...
We introduce Deep Variational Bayes Filters (DVBF), a new method for unsupervised learning and identification of latent Markovian state space models. Leveraging recent advances in Stochastic Gradient Bayes, DVBF can overcome intractable inference distributions via variational inference. Thus, it handle highly nonlinear input data with temporal spatial dependencies such as image sequences without domain knowledge. Our experiments show that enabling backpropagation through transitions enforces...
We introduce a method based on support vector machines which can detect opening and closing actions of the human thumb, index finger, other fingers recorded via surface EMG only. The is shown to be robust across sessions used independently position arm. With these stability criteria, ideally suited for control active prosthesis with high number degrees freedom. successfully demonstrated robotic four-finger hand, grasp objects
Anthropomorphic robots that aim to approach human performance agility and efficiency are typically highly redundant not only in their kinematics but also actuation. Variable-impedance actuators, used drive many of these devices, capable modulating torque impedance (stiffness and/or damping) simultaneously, continuously, independently. These actuators are, however, nonlinear assert numerous constraints, e.g., range, rate, effort limits on the dynamics. Finding a control strategy makes use...
For many tasks, tactile or visual feedback is helpful even crucial. However, designing controllers that take such high-dimensional into account non-trivial. Therefore, robots should be able to learn skills through trial and error by using reinforcement learning algorithms. The input domain for however, might include strongly correlated non-relevant dimensions, making it hard specify a suitable metric on domains. Auto-encoders specialize in finding compact representations, where defining...
Combined efforts in the fields of neuroscience, computer science and biology allowed to design biologically realistic models brain based on spiking neural networks. For a proper validation these models, an embodiment dynamic rich sensory environment, where model is exposed sensory-motor task, needed. Due complexity that, at current stage, cannot deal with real-time constraints, it not possible embed them into real world task. Rather, has be simulated as well. While adequate tools exist...
The recognition of actions from video sequences has many applications in health monitoring, assisted living, surveillance, and smart homes. Despite advances sensing, particular related to 3D video, the methodologies process data are still subject research. We demonstrate superior results by a system which combines recurrent neural networks with convolutional voting approach. gated-recurrent-unit-based particularly well-suited distinguish based on long-term information optical tracking data;...
A neural map algorithm has been employed to control a five-joint pneumatic robot arm and gripper through feedback from two video cameras. The pneumatically driven (SoftArm) in this investigation shares essential mechanical characteristics with skeletal muscle systems. To the position of arm, 200 neurons formed network representing three-dimensional workspace embedded four-dimensional system coordinates cameras, learned set pressures corresponding end effector positions, as well 3/spl times/4...
In this paper we describe and practically demonstrate a robotic arm/hand system that is controlled in real-time 6D Cartesian space through measured human muscular activity. The soft-robotics control architecture of the ensures safe physical robot interaction as well stable behaviour while operating an unstructured environment. Muscular realised via surface electromyography, non-invasive simple way to gather activity from skin. A standard supervised machine learning used create map muscle...
Nerve tissue contains a high density of chemical synapses, about 1 per µm3 in the mammalian cerebral cortex. Thus, even for small blocks nerve tissue, dense connectomic mapping requires identification millions to billions synapses. While focus data analysis has been on neurite reconstruction, synapse detection becomes limiting when datasets grow size and is required. Here, we report SynEM, method automated synapses from conventionally en-bloc stained 3D electron microscopy image stacks. The...
The development of processes and tools for ethical, trustworthy, legal AI is only beginning. At the same time, requirements are emerging in various jurisdictions, following a deluge ethical guidelines. It therefore key to explore necessary practices that must be adopted ensure quality systems, mitigate their potential risks enable compliance. Ensuring negative impacts on individuals, society, environment mitigated will depend many factors, including capacity properly regulate its deployment...
We present a setup to control four-finger anthropomorphic robot hand using dataglove. To be able accurately use the dataglove we implemented nonlinear learning calibration novel neural network technique. Experiments show that resulting positioning error not exceeding 1.8 mm, but typically 0.5 per finger can obtained; this accuracy is sufficiently precise for grasping tasks. Based on solution mapping of human and artificial workspaces enables an operator intuitively easily telemanipulate...
A novel approach to antagonism in robotic systems is introduced and investigated as the basis for an unequalled, highly anthropomorphic hand–arm system currently being developed. This system, consisting of a 19-d.o.f. hand 7-d.o.f. flexible arm, will be based on antagonistic principles order study mimic human musculoskeletal well advance safety robotics.