Search

Autonomous systems, like vehicles or robots, require reliable, accurate, fast, resource-efficient, scalable, and low-latency trajectory predictions to get initial knowledge about future locations and movements of surrounding objects for safe human-machine interaction. Furthermore, they need to know the uncertainty of the predictions for risk assessment to provide safe path planning. This paper presents a lightweight method to address these requirements, combining Long Short-Term Memory and Mixture Density Networks. Our method predicts probability distributions, including confidence level estimations for positional uncertainty to support subsequent risk management applications and runs on a low-power embedded platform. We discuss essential requirements for human trajectory prediction in autonomous vehicle applications and demonstrate our method's performance using multiple traffic-related datasets. Furthermore, we explain reliability and sharpness metrics and show how important they are to guarantee the correctness and robustness of a model's predictions and uncertainty assessments. These essential evaluations have so far received little attention for no good reason. Our approach focuses entirely on real-world applicability. Verifying prediction uncertainties and a model's reliability are central to autonomous real-world applications. Our framework and code are available at: https://github.com/kav-institute/mdn_trajectory_forecasting.

Bose-Einstein condensates (BECs) of neutral atoms constitute an important quantum system for fundamental research and precision metrology. Many applications require short preparation times of BECs, for example, for optimized data acquisition rates in scientific applications, and reduced dead times and improved bandwidths for atomic quantum sensors. Here, we report on the generation of rubidium BECs with a repetition rate of more than 2 Hz. The system relies on forced evaporation in a dynamically adjusted optical potential, which is created by the spatial modulation of laser beams. Our system provides a versatile source of the ubiquitous rubidium BECs, and promotes their exploitation for high-precision atom interferometers.

This paper extends our earlier work [arXiv:2309.13595] on the $L^p$ approximation of the shape tensor by Laurain and Sturm. In particular, it is shown that the weighted $L^p$ distance to an affine space of admissible symmetric shape tensors satisfying a divergence constraint provides the shape gradient with respect to the $L^{p^\ast}$-norm (where $1/p + 1/p^\ast = 1$) of the elastic strain associated with the shape deformation. This approach allows the combination of two ingredients which have already been used successfully in numerical shape optimization: (i) departing from the Hilbert space framework towards the Lipschitz topology approximated by $W^{1,p^\ast}$ with $p^\ast > 2$ and (ii) using the symmetric rather than the full gradient to define the norm. Similarly to [arXiv:2309.13595], the $L^p$ distance measures the shape stationarity by means of the dual norm of the shape derivative with respect to the above-mentioned $L^{p^\ast}$-norm of the elastic strain. Moreover, the Lagrange multiplier for the momentum balance constraint constitute the steepest descent deformation with respect to this norm. The finite element realization of this approach is done using the weakly symmetric PEERS element and its three-dimensional counterpart, respectively. The resulting piecewise constant approximation for the Lagrange multiplier is reconstructed to a shape gradient in $W^{1,p^\ast}$ and used in an iterative procedure towards the optimal shape.

Optimizing the injection process in particle accelerators is crucial for enhancing beam quality and operational efficiency. This paper presents a framework for utilizing Reinforcement Learning (RL) to optimize the injection process at accelerator facilities. By framing the optimization challenge as an RL problem, we developed an agent capable of dynamically aligning the beam's transverse space with desired targets. Our methodology leverages the Soft Actor-Critic algorithm, enhanced with domain randomization and dense neural networks, to train the agent in simulated environments with varying dynamics promoting it to learn a generalized robust policy. The agent was evaluated in live runs at the Cooler Synchrotron COSY and it has successfully optimized the beam cross-section reaching human operator level but in notably less time. An empirical study further validated the importance of each architecture component in achieving a robust and generalized optimization strategy. The results demonstrate the potential of RL in automating and improving optimization tasks at particle acceleration facilities., Comment: 17 pages, 6 figures

Although recent advances in higher-order Graph Neural Networks (GNNs) improve the theoretical expressiveness and molecular property predictive performance, they often fall short of the empirical performance of models that explicitly use fragment information as inductive bias. However, for these approaches, there exists no theoretic expressivity study. In this work, we propose the Fragment-WL test, an extension to the well-known Weisfeiler & Leman (WL) test, which enables the theoretic analysis of these fragment-biased GNNs. Building on the insights gained from the Fragment-WL test, we develop a new GNN architecture and a fragmentation with infinite vocabulary that significantly boosts expressiveness. We show the effectiveness of our model on synthetic and real-world data where we outperform all GNNs on Peptides and have 12% lower error than all GNNs on ZINC and 34% lower error than other fragment-biased models. Furthermore, we show that our model exhibits superior generalization capabilities compared to the latest transformer-based architectures, positioning it as a robust solution for a range of molecular modeling tasks.

Recent years have shown an increased use of silicon photo-multipliers (SiPM) in experiments as they are of reasonable cost, have relatively low power consumption and are easily available in a variety of form factors allowing for a large number of readout channels. At the same time, experiments are generating data at increasingly high rates requiring the use of more efficient readout systems. In this work, the dead time, efficiency, dynamic range, coincidence time resolution and energy resolution of five different readout systems at various stages of maturity are evaluated to determine the best system for acquiring data from a detector in a high rate experiment. Additional functionalities of the systems are also discussed., Comment: 12 pages, 5 figures

Radiofrequency-driven resonant spin rotators are routinely used as standard instruments in polarization experiments in particle and nuclear physics. Maintaining the continuous exact parametric spin-resonance condition of the equality of the spin rotator and the spin precession frequency during operation constitutes one of the challenges. We present a detailed analytic description of the impact of detuning the exact spin resonance on the vertical and the in-plane precessing components of the polarization. An important part of the formalism presented here is the consideration of experimentally relevant spin-decoherence effects. We discuss applications of the developed formalism to the interpretation of the experimental data on the novel pilot bunch approach to control the spin-resonance condition during the operation of the radiofrequency-driven Wien filter that is used as a spin rotator in the first direct deuteron electric dipole moment measurement at COSY. We emphasize the potential importance of the hitherto unexplored phase of the envelope of the horizontal polarization as an indicator of the stability of the radiofrequency-driven spin rotations in storage rings. The work presented here serves as a satellite publication to the work published concurrently on the proof of principle experiment about the so-called pilot bunch approach that was developed to provide co-magnetometry for the deuteron electric dipole moment experiment at COSY., Comment: 31 pages, 10 figures, 5 tables

In polarization experiments at storage rings, one of the challenges is to maintain the spin-resonance condition of a radio-frequency spin rotator with the spin-precessions of the orbiting particles. Time-dependent variations of the magnetic fields of ring elements lead to unwanted variations of the spin precession frequency. We report here on a solution to this problem by shielding (or masking) one of the bunches stored in the ring from the high-frequency fields of the spin rotator, so that the masked pilot bunch acts as a co-magnetometer for the other signal bunch, tracking fluctuations in the ring on a time scale of about one second. While the new method was developed primarily for searches of electric dipole moments of charged particles, it may have far-reaching implications for future spin physics facilities, such as the EIC and NICA., Comment: 5 pages, 3 figures + references + supplemental material (6 pages, 2 figures, 6 tables + references)

In the context of autonomous forklifts, ensuring non-collision during travel, pick, and place operations is crucial. To accomplish this, the forklift must be able to detect and locate areas of free space and potential obstacles in its environment. However, this is particularly challenging in highly dynamic environments, such as factory sites and production halls, due to numerous industrial trucks and workers moving throughout the area. In this paper, we present a novel method for free space detection, which consists of the following steps. We introduce a novel technique for surface normal estimation relying on spherical projected LiDAR data. Subsequently, we employ the estimated surface normals to detect free space. The presented method is a heuristic approach that does not require labeling and can ensure real-time application due to high processing speed. The effectiveness of the proposed method is demonstrated through its application to a real-world dataset obtained on a factory site both indoors and outdoors, and its evaluation on the Semantic KITTI dataset [2]. We achieved a mean Intersection over Union (mIoU) score of 50.90 % on the benchmark dataset, with a processing speed of 105 Hz. In addition, we evaluated our approach on our factory site dataset. Our method achieved a mIoU score of 63.30 % at 54 Hz

Complex inner-city junctions are among the most critical traffic areas for injury and fatal accidents. The development of highly automated driving (HAD) systems struggles with the complex and hectic everyday life within those areas. Sensor-equipped smart infrastructures, which can communicate and cooperate with vehicles, are essential to enable a holistic scene understanding to resolve occlusions drivers and vehicle perception systems for themselves can not cover. We introduce an intelligent research infrastructure equipped with visual sensor technology, located at a public inner-city junction in Aschaffenburg, Germany. A multiple-view camera system monitors the traffic situation to perceive road users' behavior. Both motorized and non-motorized traffic is considered. The system is used for research in data generation, evaluating new HAD sensors systems, algorithms, and Artificial Intelligence (AI) training strategies using real-, synthetic- and augmented data. In addition, the junction features a highly accurate digital twin. Real-world data can be taken into the digital twin for simulation purposes and synthetic data generation., Comment: IEEE International Smart Cities Conference (ISC2) 2021

Inner-city intersections are among the most critical traffic areas for injury and fatal accidents. Automated vehicles struggle with the complex and hectic everyday life within those areas. Sensor-equipped smart infrastructures, which can cooperate with vehicles, can benefit automated traffic by extending the perception capabilities of drivers and vehicle perception systems. Additionally, they offer the opportunity to gather reproducible and precise data of a holistic scene understanding, including context information as a basis for training algorithms for various applications in automated traffic. Therefore, we introduce the Infrastructural Multi-Person Trajectory and Context Dataset (IMPTC). We use an intelligent public inner-city intersection in Germany with visual sensor technology. A multi-view camera and LiDAR system perceives traffic situations and road users' behavior. Additional sensors monitor contextual information like weather, lighting, and traffic light signal status. The data acquisition system focuses on Vulnerable Road Users (VRUs) and multi-agent interaction. The resulting dataset consists of eight hours of measurement data. It contains over 2,500 VRU trajectories, including pedestrians, cyclists, e-scooter riders, strollers, and wheelchair users, and over 20,000 vehicle trajectories at different day times, weather conditions, and seasons. In addition, to enable the entire stack of research capabilities, the dataset includes all data, starting from the sensor-, calibration- and detection data until trajectory and context data. The dataset is continuously expanded and is available online for non-commercial research at https://github.com/kav-institute/imptc-dataset., Comment: IEEE Intelligent Vehicles Conference (IV) 2023

Objective. Prompt-gamma imaging encompasses several approaches for online monitoring of beam range or deposited dose distribution in proton therapy. We test one of the imaging techniques - a coded mask approach - both experimentally and via simulations. Approach. Two imaging setups have been investigated experimentally. Each of them comprised a structured tungsten collimator in a form of a MURA mask and a LYSO:Ce scintillation detector of fine granularity. The setups differed in the detector dimensions and the operation mode (1D or 2D imaging). A series of measurements with radioactive sources have been conducted, testing the setups' performance of near-field gamma imaging. Additionally, Monte Carlo simulations of a larger setup of the same type were conducted, investigating its performance with a realistic gamma source distribution occurring during proton therapy. Main results. The images of point-like sources reconstructed from two smallscale prototypes' data using the MLEM algorithm constitute the experimental proof of principle for the near-field coded-mask imaging modality, both in the 1D and the 2D mode. Their precision allowed us to calibrate out certain systematic offsets appearing due to the misalignment of setup elements. The simulation of the full-scale setup yielded a mean distal falloff retrieval precision of 0.72 mm in the studies for beam energy range 89.5-107.9 MeV and with 1x10^8 protons (typical number for single distal spots). The implemented algorithm of image reconstruction is relatively fast - a typical procedure needs several seconds. Significance. Coded-mask imaging appears a valid option for proton therapy monitoring. The results of simulations let us conclude that the proposed fullscale setup is competitive to the knife-edge-shaped and the multiparalell slit cameras investigated by other groups.

Space-based stellar coronagraph instruments aim to directly image exoplanets that are a fraction of an arcsecond separation and ten billion times fainter than their host star. To achieve this, one or more deformable mirrors (DMs) are used in concert with coronagraph masks to control the wavefront and minimize diffracted starlight in a region of the image known as the ``dark zone" or ``dark hole." The DMs must have a high number of actuators (50 to 96 across) to allow dark holes that are large enough to image a range of desired exoplanet separations. In addition, the surfaces of the DMs must be controlled at the picometer level to enable the required contrast. Any defect in the mechanical structure of the DMs or electronic system could significantly impact the scientific potential of the mission. Thus, NASA's Exoplanet Exploration Program (ExEP) procured two 50$\times$50 microelectromechanical (MEMS) DMs manufactured by Boston Micromachines Corporation (BMC) to test their robustness to the vibrational environment that the DMs will be exposed to during launch. The DMs were subjected to a battery of functional and high-contrast imaging tests before and after exposure to flight-like random vibrations. The DMs did not show any significant functional nor performance degradation at $10^{-8}$ contrast levels., Comment: Accepted for publication in JATIS

Recent advances in machine learning for molecules exhibit great potential for facilitating drug discovery from in silico predictions. Most models for molecule generation rely on the decomposition of molecules into frequently occurring substructures (motifs), from which they generate novel compounds. While motif representations greatly aid in learning molecular distributions, such methods struggle to represent substructures beyond their known motif set. To alleviate this issue and increase flexibility across datasets, we propose MAGNet, a graph-based model that generates abstract shapes before allocating atom and bond types. To this end, we introduce a novel factorisation of the molecules' data distribution that accounts for the molecules' global context and facilitates learning adequate assignments of atoms and bonds onto shapes. Despite the added complexity of shape abstractions, MAGNet outperforms most other graph-based approaches on standard benchmarks. Importantly, we demonstrate that MAGNet's improved expressivity leads to molecules with more topologically distinct structures and, at the same time, diverse atom and bond assignments.

In this work, we propose an extension of conventional image data by an additional channel in which the associated projection properties are encoded. This addresses the issue of sensor-dependent object representation in projection-based sensors, such as LiDAR, which can lead to distorted physical and geometric properties due to variations in sensor resolution and field of view. To that end, we propose an architecture for processing this data in an instance segmentation framework. We focus specifically on LiDAR as a key sensor modality for machine vision tasks and highly automated driving (HAD). Through an experimental setup in a controlled synthetic environment, we identify a bias on sensor resolution and field of view and demonstrate that our proposed method can reduce said bias for the task of LiDAR instance segmentation. Furthermore, we define our method such that it can be applied to other projection-based sensors, such as cameras. To promote transparency, we make our code and dataset publicly available. This method shows the potential to improve performance and robustness in various machine vision tasks that utilize projection-based sensors.

Machine learning for molecules holds great potential for efficiently exploring the vast chemical space and thus streamlining the drug discovery process by facilitating the design of new therapeutic molecules. Deep generative models have shown promising results for molecule generation, but the benefits of specific inductive biases for learning distributions over small graphs are unclear. Our study aims to investigate the impact of subgraph structures and vocabulary design on distribution learning, using small drug molecules as a case study. To this end, we introduce Subcover, a new subgraph-based fragmentation scheme, and evaluate it through a two-step variational auto-encoder. Our results show that Subcover's improved identification of chemically meaningful subgraphs leads to a relative improvement of the FCD score by 30%, outperforming previous methods. Our findings highlight the potential of Subcover to enhance the performance and scalability of existing methods, contributing to the advancement of drug discovery., Comment: Accepted for publication at the MLDD workshop, ICLR 2023

The complex non-linear processes in multi-dimensional parameter spaces, that are typical for an accelerator, are a natural application for machine learning algorithms. This paper reports on the use of Bayesian optimization for the optimization of the Injection Beam Line (IBL) of the Cooler Synchrotron storage ring COSY at the Forschungszentrum J\"ulich, Germany. Bayesian optimization is a machine learning method that optimizes a continuous objective function using limited observations. The IBL is composed of 15 quadrupoles and 28 steerers. The goal is to increase the beam intensity inside the storage ring. The results showed the effectiveness of the Bayesian optimization in achieving better/faster results compared to manual optimization.

High data acquisition rates and low-noise detection of ultracold neutral atoms present important challenges for the state tomography and interferometric application of entangled quantum states in Bose-Einstein condensates. In this article, we present a high-flux source of $^{87}$Rb Bose-Einstein condensates combined with a number-resolving detection. We create Bose-Einstein condensates of $2\times10^5$ atoms with no discernible thermal fraction within $3.3$ s using a hybrid evaporation approach in a magnetic/optical trap. For the high-fidelity tomography of many-body quantum states in the spin degree of freedom [arXiv:2207.01270], it is desirable to select a single mode for a number-resolving detection. We demonstrate the low-noise selection of subsamples of up to $16$ atoms and their subsequent detection with a counting noise below $0.2$ atoms. The presented techniques offer an exciting path towards the creation and analysis of mesoscopic quantum states with unprecedented fidelities, and their exploitation for fundamental and metrological applications., Comment: Corrected figures, updated references

Single-cell reference atlases are large-scale, cell-level maps that capture cellular heterogeneity within an organ using single cell genomics. Given their size and cellular diversity, these atlases serve as high-quality training data for the transfer of cell type labels to new datasets. Such label transfer, however, must be robust to domain shifts in gene expression due to measurement technique, lab specifics and more general batch effects. This requires methods that provide uncertainty estimates on the cell type predictions to ensure correct interpretation. Here, for the first time, we introduce uncertainty quantification methods for cell type classification on single-cell reference atlases. We benchmark four model classes and show that currently used models lack calibration, robustness, and actionable uncertainty scores. Furthermore, we demonstrate how models that quantify uncertainty are better suited to detect unseen cell types in the setting of atlas-level cell type transfer., Comment: Workshop paper at the 2022 ICML Workshop on Computational Biology

Based on the notion that the local dark-matter field of axions or axion-like particles (ALPs) in our Galaxy induces oscillating couplings to the spins of nucleons and nuclei (via the electric dipole moment of the latter and/or the paramagnetic axion-wind effect), we establish the feasibility of a new method to search for ALPs in storage rings. Based on previous work that allows us to maintain the in-plane polarization of a stored deuteron beam for a few hundred seconds, we performed a first proof-of-principle experiment at the Cooler Synchrotron COSY to scan momenta near 970 MeV/c. This entailed a scan of the spin precession frequency. At resonance between the spin precession frequency of deuterons and the ALP-induced EDM oscillation frequency there will be an accumulation of the polarization component out of the ring plane. Since the axion frequency is unknown, the momentum of the beam and consequently the spin precession frequency were ramped to search for a vertical polarization change that would occur when the resonance is crossed. At COSY, four beam bunches with different polarization directions were used to make sure that no resonance was missed because of the unknown relative phase between the polarization precession and the axion/ALP field. A frequency window of 1.5-kHz width around the spin precession frequency of 121 kHz was scanned. We describe the experimental procedure and a test of the methodology with the help of a radiofrequency Wien filter located on the COSY ring. No ALP resonance was observed. As a consequence an upper limit of the oscillating EDM component of the deuteron as well as its axion coupling constants are provided., Comment: 27 pages, 24 figures, 7 tables, 104 references

The high-fidelity analysis of many-body quantum states of indistinguishable atoms requires the accurate counting of atoms. Here we report the tomographic reconstruction of an atom-number-resolving detector. The tomography is performed with an ultracold rubidium ensemble that is prepared in a coherent spin state by driving a Rabi coupling between the two hyperfine clock levels. The coupling is followed by counting the occupation number in one level. We characterize the fidelity of our detector and show that a negative-valued Wigner function is associated with it. Our results offer an exciting perspective for the high-fidelity reconstruction of entangled states and can be applied for a future demonstration of Heisenberg-limited atom interferometry.