检索结果-内蒙古大学图书馆

Federated Learning for Privacy-Preserving Feedforward control in Multi-Agent Systems

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Weber, Jakob Gurtner, Markus Alt, Benedikt Trachte, Adrian Kugi, Andreas The Center for Vision Automation & Control AIT Austrian Institute of Technology GmbH Vienna Austria The Robert Bosch GmbH Renningen Germany The Automation and Control Institute TU Wien Austria The AIT Austrian Institute of Technology GmbH Vienna Austria

Feedforward control (FF) is often combined with feedback control (FB) in many control systems, improving tracking performance, efficiency, and stability. However, designing effective data-driven FF controllers in multi-agent systems requires significant data collection, including transferring private or proprietary data, which raises privacy concerns and incurs high communication costs. Therefore, we propose a novel approach integrating Federated Learning (FL) into FF control to address these challenges. This approach enables privacy-preserving, communication-efficient, and decentralized continuous improvement of FF controllers across multiple agents without sharing personal or proprietary data. By leveraging FL, each agent learns a local, neural FF controller using its data and contributes only model updates to a global aggregation process, ensuring data privacy and scalability. We demonstrate the effectiveness of our method in an autonomous driving use case. Therein, vehicles equipped with a trajectory-tracking feedback controller are enhanced by FL-based neural FF control. Simulations highlight significant improvements in tracking performance compared to pure FB control, analogous to model-based FF control. We achieve comparable tracking performance without exchanging private vehicle-specific data compared to a centralized neural FF control. Our results underscore the potential of FL-based neural FF control to enable privacy-preserving learning in multi-agent control systems, paving the way for scalable and efficient autonomous systems applications. © 2025, CC BY-NC-ND.

关键词： Differential privacy

Photometric visibility matrix for the automatic selection of optimal viewpoints

学校读者我要写书评

暂无评论

Photometric visibility matrix for the automatic selection of...

International Conference on 3D Imaging, Modeling, Processing, Visualization and Transmission (3DIMPVT)

作者： Vanessa Staderini Tobias Glück Roberto Mecca Petra Gospodnetic Philipp Schneider Andreas Kugi Center for Vision Automation & Control AIT Austrian Institute of Technology GmbH Vienna Austria Image Processing Department Fraunhofer ITWM Kaiserslautern Germany

ISBN: (数字)9798350362459

ISBN: (纸本)9798350362466

Automated visual quality inspection is a core topic of robotics and computer vision. In industrial applications, the CAD model of the object to be inspected is often known and can be used to generate appropriate sensor poses (viewpoints) from which to inspect the object’s surface and assure the quality of its geometry. Current approaches in this field generate optimal viewpoints by evaluating the geometric coverage but the photometric appearance of the object is usually not considered. This lack of photometric information results in a loss of crucial cues to establish actual visibility, especially when the object to inspect presents specular highlights (e.g., polished metal parts) and a complex geometry. In this paper, we propose integrating photometric information into the viewpoint evaluation to consider the object’s appearance. To achieve this, we embed a bidirectional reflectance distribution function (BRDF) within the evaluation of the viewpoint candidates. We benchmark different BRDFs with increasingly realistic rendering to prove the concept of our approach. Specifically, we consider the Blinn-Phong and Cook-Torrance reflectance models. Our simulation results demonstrate the suitability and importance of using a photometric approach that considers material properties and selects optimal viewpoints for specific materials.

关键词： Geometry Reflectivity Solid modeling Visualization Three-dimensional displays Service robots Simulation

Combining Federated Learning and control: A Survey

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Weber, Jakob Gurtner, Markus Lobe, Amadeus Trachte, Adrian Kugi, Andreas The Center for Vision Automation & Control AIT Austrian Institute of Technology GmbH Vienna Austria The Robert Bosch GmbH Renningen Germany The Automation and Control Institute TU Wien Austria The AIT Austrian Institute of Technology GmbH Vienna Austria

This survey provides an overview of combining Federated Learning (FL) and control to enhance adaptability, scalability, generalization, and privacy in (nonlinear) control applications. Traditional control methods rely on controller design models, but real-world scenarios often require online model retuning or learning. FL offers a distributed approach to model training, enabling collaborative learning across distributed devices while preserving data privacy. By keeping data localized, FL mitigates concerns regarding privacy and security while reducing network bandwidth requirements for communication. This survey summarizes the state-of-the-art concepts and ideas of combining FL and control. The methodical benefits are further discussed, culminating in a detailed overview of expected applications, from dynamical system modeling over controller design, focusing on adaptive control, to knowledge transfer in multi-agent decision-making systems. © 2024, CC BY.

关键词： Data privacy

GPU-Accelerated Motion Planning of an Underactuated Forestry Crane in Cluttered Environments

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Vu, Minh Nhat Ebmer, Gerald Watcher, Alexander Ecker, Marc-Philip Nguyen, Giang Glueck, Tobias TU Wien Vienna Austria Center for Vision Automation & Control AIT Austrian Institute of Technology GmbH Vienna Austria Institute for Artificial Intelligence University of Bremen Germany

Autonomous large-scale machine operations require fast, efficient, and collision-free motion planning while addressing unique challenges such as hydraulic actuation limits and underactuated joint dynamics. This paper presents a novel two-step motion planning framework designed for an underactuated forestry crane. The first step employs GPU-accelerated stochastic optimization to rapidly compute a globally shortest collision-free path. The second step refines this path into a dynamically feasible trajectory using a trajectory optimizer that ensures compliance with system dynamics and actuation constraints. The proposed approach is benchmarked against conventional techniques, including RRT-based methods and purely optimization-based approaches. Simulation results demonstrate substantial improvements in computation speed and motion feasibility, making this method highly suitable for complex crane systems. © 2025, CC BY-NC-ND.

关键词： Stochastic systems

Curiosity-Driven Imagination: Discovering Plan Operators and Learning Associated Policies for Open-World Adaptation

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Lorang, Pierrick Lu, Hong Scheutz, Matthias Tufts University 419 Boston Ave MedfordMA02155 United States AIT Austrian Institute of Technology GmbH Center for Vision Automation & Control Vienna Austria

Adapting quickly to dynamic, uncertain environments—often called "open worlds"—remains a major challenge in robotics. Traditional Task and Motion Planning (TAMP) approaches struggle to cope with unforeseen changes, are data-inefficient when adapting, and do not leverage world models during learning. We address this issue with a hybrid planning and learning system that integrates two models: a low-level neural network-based model that learns stochastic transitions and drives exploration via an Intrinsic Curiosity Module (ICM), and a high-level symbolic planning model that captures abstract transitions using operators, enabling the agent to plan in an "imaginary" space and generate reward machines. Our evaluation in a robotic manipulation domain with sequential novelty injections demonstrates that our approach converges faster and outperforms state-of-the-art hybrid methods. Copyright © 2025, The Authors. All rights reserved.

关键词： Stochastic systems

Three-Dimensional Waypoint Navigation of Multicopters by Attitude and Throttle Commands using Off-Policy Reinforcement Learning

学校读者我要写书评

暂无评论

Three-Dimensional Waypoint Navigation of Multicopters by Att...

International Conference on Unmanned Aircraft Systems (ICUAS)

作者： Francesco d’ Apolito Christoph Sulzbachner Center for Vision Automation and Control AIT - Austrian Institute of Technology Vienna Austria

ISBN: (数字)9781665405935

ISBN: (纸本)9781665405942

Artificial intelligence, in particular machine learning, is becoming increasingly important in automation and robotics. Machine learning approaches are also becoming more and more accepted in aviation. In particular, Reinforcement Learning is gaining more attention in navigation and control problems, for example in training flight manoeuvres. This paper aims to investigate the use of Off-Policy Reinforcement Learning techniques for three-dimensional waypoint navigation of multicopters by providing roll, pitch and throttle commands. It describes and compare the trainings performed using two well-known Off-Policy algorithms, namely the Deep Deterministic Policy Gradient (DDPG) and the Soft Actor Critic (SAC). Furthermore, we investigate the impact of the reward definition on the training outcome. For each of the used algorithm, two agents are trained with two different reward definitions. Finally, the paper shows the validations performed to evaluate the performance of the four trained agents under different known and unknown conditions. Their performances are evaluated and compared with respect to the training algorithm and the reward definition used.

关键词： Training automation Navigation Reinforcement learning Aircraft Robots

Design and analysis of a class of planar cable-driven parallel robots with arbitrary rotation

学校读者我要写书评

暂无评论

IFAC-PapersOnLine 2022年第27期55卷 82-88页

作者： Georg Feiler Michael Schwegel Ulrich Knechtelsdorfer Andreas Kugi Automation and Control Institute (ACIN) TU Wien Gußhausstraße 27-29 1040 Wien Center for Vision Automation & Control AIT Austrian Institute of Technology 1210 Wien

This paper presents and analyzes a novel class of planar cable-driven parallel robots. The design allows an arbitrary rotation of the end-effector by wrapping the cables on a cylindrical fixture on the end-effector platform. In contrast to the state of the art, this results in a simple and robust design without additional moving parts attached to the end effector. Based on Burnside's lemma known from group theory, the set of all possible design configurations for a fixed number of cables is derived. Moreover, the workspace is analyzed by means of the classical manipulability measure known from the literature. In order to also capture the minimum tension in the cables and the directional dependence of the achievable forces, which are determining factors for the workspace, a new force manipulability measure is introduced. This yields results which are related to the well-known wrench-feasible workspace. The proposed concepts are demonstrated by two different designs for planar cable-driven parallel robots with 4 and 6 cables.

关键词： cable-driven parallel robot kinematic robot design manipulability workspace analysis

Nonlinear feedback stabilisation and stochastic disturbance suppression of actively Q-switched lasers

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Tarra, Lukas Deutschmann-Olek, Andreas Kugi, Andreas Automation and Control Institute TU Wien Gußhausstraße 27-29 Vienna1040 Austria Center for Vision Automation & Control Austrian Institute of Technology Giefinggasse 4 Vienna1210 Austria

Actively Q-switched lasers are widely used tools which are required to produce stable output pulse energies for many applications. In this paper, a model-based control concept for actively Q-switched lasers is presented which stabilises their nonlinear pulse-to-pulse dynamics and rejects stochastic disturbances arising from amplified spontaneous emission. The feasibility of the control task is demonstrated to strongly depend on the design of the semi-active prelasing approach. In contrast to state-of-the-art hardware-based controllers, the proposed concept is flexible and cost-effective as it is not tailored to specific operation parameters. © 2023, CC BY.

关键词： Stochastic systems

PDAVIS: Bio-inspired Polarization Event Camera

学校读者我要写书评

暂无评论

PDAVIS: Bio-inspired Polarization Event Camera

2023 IEEE/CVF Conference on Computer vision and Pattern Recognition Workshops, CVPRW 2023

作者： Haessig, Germain Joubert, Damien Haque, Justin Milde, Moritz B. Delbruck, Tobi Gruev, Viktor Center for Vision Automation & Control High-Performance Vision Systems AIT Austrian Institute of Technology Vienna Austria University of Zurich Institute of Neuroinformatics ETH Zurich Switzerland Western Sydney University International Centre for Neuromorphic Systems The MARCS Institute Sydney Australia University of Illinois at Urbana-Champaign Department of Electrical and Computer Engineering UrbanaIL United States

ISBN: (纸本)9798350302493

The stomatopod (mantis shrimp) visual system has recently provided a blueprint for the design of paradigm-shifting polarization and multispectral imaging sensors, enabling solutions to challenging medical and remote sensing problems. However, these bioinspired frame-based cameras lack the high dynamic range and asynchronous polarization vision capabilities of the stomatopod visual system, limiting temporal resolution to ~12ms and dynamic range to ~72dB. Here we present a novel stomatopod-inspired polarization camera which mimics the sustained and transient biological visual pathways to save power and sample data beyond the maximum Nyquist frame rate. This bio-inspired sensor simultaneously captures both synchronous intensity frames and asynchronous polarization brightness change information with submillisecond latencies over a millionfold range of illumination. Our PDAVIS camera is comprised of 346x260 pixels, organized in 2-by-2 macropixels, which filter the incoming light with four linear polarization filters offset by 45°. Polarization information is reconstructed using both low-cost and low-latency event-based algorithms and more accurate but slower deep neural networks. Our sensor is used to image high dynamic range polarization scenes that vary at high speeds and to observe the dynamical properties of single collagen fibers in a bovine tendon under rapid cyclical ***: https://***/mFuCeTMWEqY © 2023 IEEE.

关键词： Polarization