Showing 1–7 of 7 results for author: Trost, F

Human-Robot Gym: Benchmarking Reinforcement Learning in Human-Robot Collaboration

Authors: Jakob Thumm, Felix Trost, Matthias Althoff

Abstract: Deep reinforcement learning (RL) has shown promising results in robot motion planning with first attempts in human-robot collaboration (HRC). However, a fair comparison of RL approaches in HRC under the constraint of guaranteed safety is yet to be made. We, therefore, present human-robot gym, a benchmark suite for safe RL in HRC. Our benchmark suite provides eight challenging, realistic HRC tasks… ▽ More Deep reinforcement learning (RL) has shown promising results in robot motion planning with first attempts in human-robot collaboration (HRC). However, a fair comparison of RL approaches in HRC under the constraint of guaranteed safety is yet to be made. We, therefore, present human-robot gym, a benchmark suite for safe RL in HRC. Our benchmark suite provides eight challenging, realistic HRC tasks in a modular simulation framework. Most importantly, human-robot gym includes a safety shield that provably guarantees human safety. We are, thereby, the first to provide a benchmark suite to train RL agents that adhere to the safety specifications of real-world HRC. This bridges a critical gap between theoretic RL research and its real-world deployment. Our evaluation of six tasks led to three key results: (a) the diverse nature of the tasks offered by human-robot gym creates a challenging benchmark for state-of-the-art RL methods, (b) incorporating expert knowledge in RL training in the form of an action-based reward can outperform the expert, and (c) our agents negligibly overfit to training data. △ Less

Submitted 25 June, 2024; v1 submitted 9 October, 2023; originally announced October 2023.

Decreasing ultrafast X-ray pulse durations with saturable absorption and resonant transitions

Authors: Sebastian Cardoch, Fabian Trost, Howard A. Scott, Henry N. Chapman, Carl Caleman, Nicusor Timneanu

Abstract: Saturable absorption is a nonlinear effect where a material's ability to absorb light is frustrated due to a high influx of photons and the creation of electron vacancies. Experimentally induced saturable absorption in copper revealed a reduction in the temporal duration of transmitted X-ray laser pulses, but a complete understanding of this process is still missing. In this computational work, we… ▽ More Saturable absorption is a nonlinear effect where a material's ability to absorb light is frustrated due to a high influx of photons and the creation of electron vacancies. Experimentally induced saturable absorption in copper revealed a reduction in the temporal duration of transmitted X-ray laser pulses, but a complete understanding of this process is still missing. In this computational work, we employ non-local thermodynamic equilibrium plasma simulations to study the interaction of femtosecond X-rays and copper. Following the onset of frustrated absorption, we find that a $K\text{--}M$ resonant transition occurring at highly charged states turns copper opaque again. The changes in absorption generate a transient transparent window responsible for the shortened transmission signal. We also propose using fluorescence induced by the incident beam as an alternative source to achieve shorter X-ray pulses. Intense femtosecond X-ray pulses are valuable to probe the structure and dynamics of biological samples or to reach extreme states of matter. Shortened pulses could be relevant for emerging imaging techniques. △ Less

Submitted 10 October, 2022; originally announced October 2022.

Journal ref: Phys. Rev. E 107, 015205 2023

Control of Two-way Coupled Fluid Systems with Differentiable Solvers

Authors: Brener Ramos, Felix Trost, Nils Thuerey

Abstract: We investigate the use of deep neural networks to control complex nonlinear dynamical systems, specifically the movement of a rigid body immersed in a fluid. We solve the Navier Stokes equations with two way coupling, which gives rise to nonlinear perturbations that make the control task very challenging. Neural networks are trained in an unsupervised way to act as controllers with desired charact… ▽ More We investigate the use of deep neural networks to control complex nonlinear dynamical systems, specifically the movement of a rigid body immersed in a fluid. We solve the Navier Stokes equations with two way coupling, which gives rise to nonlinear perturbations that make the control task very challenging. Neural networks are trained in an unsupervised way to act as controllers with desired characteristics through a process of learning from a differentiable simulator. Here we introduce a set of physically interpretable loss terms to let the networks learn robust and stable interactions. We demonstrate that controllers trained in a canonical setting with quiescent initial conditions reliably generalize to varied and challenging environments such as previously unseen inflow conditions and forcing, although they do not have any fluid information as input. Further, we show that controllers trained with our approach outperform a variety of classical and learned alternatives in terms of evaluation metrics and generalization capabilities. △ Less

Submitted 1 June, 2022; originally announced June 2022.

Comments: 24 pages, 18 figures

MSC Class: 68T07

On the use of multilayer Laue lenses with X-ray Free Electron Lasers

Authors: Mauro Prasciolu, Kevin T. Murray, Nikolay Ivanov, Holger Fleckenstein, Martin Domaracký, Luca Gelisio, Fabian Trost, Kartik Ayyer, Dietrich Krebs, Steve Aplin, Salah Awel, Ulrike Boesenberg, Anton Barty, Armando D. Estillore, Matthias Fuchs, Yaroslav Gevorkov, Joerg Hallmann, Chan Kim, Juraj Knoška, Jochen Küpper, Chufeng Li, Wei Lu, Valerio Mariani, Andrew J. Morgan, Johannes Möller , et al. (12 additional authors not shown)

Abstract: Multilayer Laue lenses were used for the first time to focus x-rays from an X-ray Free Electron Laser (XFEL). In an experiment, which was performed at the European XFEL, we demonstrated focusing to a spot size of a few tens of nanometers. A series of runs in which the number of pulses per train was increased from 1 to 2, 3, 4, 5, 6, 7, 10, 20 and 30 pulses per train, all with a pulse separation of… ▽ More Multilayer Laue lenses were used for the first time to focus x-rays from an X-ray Free Electron Laser (XFEL). In an experiment, which was performed at the European XFEL, we demonstrated focusing to a spot size of a few tens of nanometers. A series of runs in which the number of pulses per train was increased from 1 to 2, 3, 4, 5, 6, 7, 10, 20 and 30 pulses per train, all with a pulse separation of 3.55 us, was done using the same set of lenses. The increase in the number of pulses per train was accompanied with an increase of x-ray intensity (transmission) from 9% to 92% at 5 pulses per train, and then the transmission was reduced to 23.5 % when the pulses were increased further. The final working condition was 30 pulses per train and 23.5% transmission. Only at this condition we saw that the diffraction efficiency of the MLLs changed over the course of a pulse train, and this variation was reproducible from train to train. We present the procedure to align and characterize these lenses and discuss challenges working with the pulse trains from this unique x-ray source. △ Less

Submitted 22 March, 2022; originally announced March 2022.

Physics-based Deep Learning

Authors: Nils Thuerey, Philipp Holl, Maximilian Mueller, Patrick Schnell, Felix Trost, Kiwon Um

Abstract: This digital book contains a practical and comprehensive introduction of everything related to deep learning in the context of physical simulations. As much as possible, all topics come with hands-on code examples in the form of Jupyter notebooks to quickly get started. Beyond standard supervised learning from data, we'll look at physical loss constraints, more tightly coupled learning algorithms… ▽ More This digital book contains a practical and comprehensive introduction of everything related to deep learning in the context of physical simulations. As much as possible, all topics come with hands-on code examples in the form of Jupyter notebooks to quickly get started. Beyond standard supervised learning from data, we'll look at physical loss constraints, more tightly coupled learning algorithms with differentiable simulations, as well as reinforcement learning and uncertainty modeling. We live in exciting times: these methods have a huge potential to fundamentally change what computer simulations can achieve. △ Less

Submitted 25 April, 2022; v1 submitted 11 September, 2021; originally announced September 2021.

Comments: PBDL v0.2, available online at: https://www.physicsbaseddeeplearning.org/

Photoelectron spectroscopy of laser-dressed atomic helium

Authors: Severin Meister, Aaron Bondy, Kirsten Schnorr, Sven Augustin, Hannes Lindenblatt, Florian Trost, Xinhua Xie, Markus Braune, Rolf Treusch, Nicolas Douguet, Thomas Pfeifer, Klaus Bartschat, Robert Moshammer

Abstract: Photoelectron emission from excited states of laser-dressed atomic helium is analyzed with respect to laser intensity-dependent excitation energy shifts and angular distributions. In the two-color XUV (exteme ultra\-violet) -- IR (infrared) measurement, the XUV photon energy is scanned between \SI{20.4}{\electronvolt} and the ionization threshold at \SI{24.6}{\electronvolt}, revealing electric dip… ▽ More Photoelectron emission from excited states of laser-dressed atomic helium is analyzed with respect to laser intensity-dependent excitation energy shifts and angular distributions. In the two-color XUV (exteme ultra\-violet) -- IR (infrared) measurement, the XUV photon energy is scanned between \SI{20.4}{\electronvolt} and the ionization threshold at \SI{24.6}{\electronvolt}, revealing electric dipole-forbidden transitions for a temporally overlap** IR pulse ($\sim\!\SI{e12}{\watt\per \centi\meter\squared}$). The interpretation of the experimental results is supported by numerically solving the time-dependent Schrödinger equation in a single-active-electron approximation. △ Less

Submitted 14 September, 2020; originally announced September 2020.

Comments: 6 pages, 5 figures

Journal ref: Phys. Rev. A 102, 062809 (2020)

Photon statistics and signal to noise ratio for incoherent diffraction imaging

Authors: Fabian Trost, Kartik Ayyer, Henry Chapman

Abstract: Intensity interferometry is a well known method in astronomy. Recently, a related method called incoherent diffractive imaging (IDI) was proposed to apply intensity correlations of x-ray fluorescence radiation to determine the 3D arrangement of the emitting atoms in a sample. Here we discuss inherent sources of noise affecting IDI and derive a model to estimate the dependence of the signal to nois… ▽ More Intensity interferometry is a well known method in astronomy. Recently, a related method called incoherent diffractive imaging (IDI) was proposed to apply intensity correlations of x-ray fluorescence radiation to determine the 3D arrangement of the emitting atoms in a sample. Here we discuss inherent sources of noise affecting IDI and derive a model to estimate the dependence of the signal to noise ratio (SNR) on the photon counts per pixel, the temporal coherence (or number of modes), and the shape of the imaged object. Simulations in two- and three-dimensions have been performed to validate the predictions of the model. We find that contrary to coherent imaging methods, higher intensities and higher detected counts do not always correspond to a larger SNR. Also, larger and more complex objects generally yield a poorer SNR despite the higher measured counts. The framework developed here should be a valuable guide to future experimental design. △ Less

Submitted 7 May, 2020; originally announced May 2020.

Journal ref: New J. Phys. 22 083070 (2020)