Geometric (India)

<div class="section abstract"><div class="htmlview paragraph">Implementation calibration of automotive radar systems plays a fundamental but crucial role to guarantee sensor performance. The commonly used method relies on the environment such as a specific test station for static calibration or a straight metal guardrail for dynamic calibration. In this paper, a sequential method for estimating the radar angle misalignment derived from the Lagrange Multiplier Method in solving an optimization problem is proposed. The sequential method, which requires radar measurements and vehicle speed measurements as input, is more environment-free and can yield a consistent estimation. A simulation study is conducted to validate the consistency and analyze the influence of noise. The result shows that the radar azimuth measurement noise has little influence that the bias could be compensated and the effect of non-gaussianity is negligible. The radar velocity measurement noise bias and vehicle speed measurement noise bias have a linear effect whose coefficient depends on the radar orientation on the angle misalignment estimation. A real road testing is carried out and the result demonstrates that the method proposed could provide an estimation with an error no more than 0.26 degrees. An artificial error with 1 degree is added to the calibration result of the proposed method to study the miscalibration influence on sensor fusion based on the neural net. The result shows that the precision and recall deteriorate significantly with the artificial error, especially for the small targets like pedestrians, the reduction could reach 14% for precision and 17% for recall.</div></div>

<div class="section abstract"><div class="htmlview paragraph">Many learning-based methods estimate ego-motion using visual sensors. However, visual sensors are prone to intense lighting variations and textureless scenarios. 4D radar, an emerging automotive sensor, complements visual sensors effectively due to its robustness in adverse weather and lighting conditions. This paper presents an end-to-end 4D radar-visual odometry (4DRVO) approach that combines sparse point cloud data from 4D radar with image information from cameras. Using the Feature Pyramid, Pose Warping, and Cost Volume (PWC) network architecture, we extract 4D radar point features and image features at multiple scales. We then employ a hierarchical iterative refinement approach to supervise the estimated pose. We propose a novel Cross-Modal Transformer (CMT) module to effectively fuse the 4D radar point modality, image modality, and 4D radar point-image connection modality at multiple scales, achieving cross-modal feature interaction and multi-modal feature fusion. Additionally, we designed a point confidence estimation module to mitigate the impact of dynamic objects on odometry estimation. Extensive experiments were conducted on the View-of-Delft (VoD) dataset, showcasing the remarkable performance and effectiveness of the proposed 4D radar-visual odometry method.</div></div>

In this paper, an approach for generalizing the Gromov-Hausdorff metric is presented, which applies to metric spaces equipped with some additional structure. A special case is the Gromov-Hausdorff-Prokhorov metric between measured metric spaces. This abstract framework unifies several existing Gromov-Hausdorff-type metrics for metric spaces equipped with a measure, a point, a closed subset, a curve, a tuple of such structures, etc. Along with reviewing these special cases in the literature, several new examples are also presented. Two frameworks are provided, one for compact metric spaces and the other for boundedly-compact pointed metric spaces. In both cases, a Gromov-Hausdorff-type metric is defined and its topological properties are studied. In particular, completeness and separability is proved under some conditions. This enables one to study random metric spaces equipped with additional structures, which is the main motivation of this work.

State-of-the-art maximum entropy models for texture synthesis are built from statistics relying on image representations defined by convolutional neural networks (CNN). Such representations capture rich structures in texture images, outperforming wavelet-based representations in this regard. However, conversely to neural networks, wavelets offer meaningful representations, as they are known to detect structures at multiple scales (e.g. edges) in images. In this work, we propose a family of statistics built upon non-linear wavelet based representations, that can be viewed as a particular instance of a one-layer CNN, using a generalized rectifier non-linearity. These statistics significantly improve the visual quality of previous classical wavelet-based models, and allow one to produce syntheses of similar quality to state-of-the-art models, on both gray-scale and color textures.

We study the performance of stochastic matching models with general compatibility graphs. Items of different classes arrive to the system according to independent Poisson processes. Upon arrival, an item is matched with a compatible item according to the First Come First Matched discipline and both items leave the system immediately. If there are no compatible items, the new arrival joins the queue of unmatched items of the same class. Compatibilities between item classes are defined by a connected graph, where nodes represent the classes of items and the edges the compatibilities between item classes. We show that such a model may exhibit a non intuitive behavior: increasing the matching flexibility by adding new edges in the matching graph may lead to a larger average population at the steady state. This performance paradox can be viewed as an analog of the Braess paradox. We show sufficient conditions for the existence or non-existence of this paradox. This performance paradox in matching models appears when specific independent sets are in saturation, i.e., the system is close to the stability condition.

Summary Field-wide seafloor subsidence has been monitored at ten hydrocarbon fields offshore Norway. Subsidence provides lateral information on compressibility and pressure depletion in the reservoirs. This has been used to calibrate geomechanical and reservoir flow models, to map reservoir and aquifer compaction and identify poorly drained compartments. Water pressure measurements have given survey-internal precision of 2–5 mm. Tidal variations in the seafloor pressure have been carefully measured and corrected, which requires a spatial grid of reference sensors. A segment of the Midgard gas field showed significantly less subsidence than expected. This was interpreted as being caused by less depletion. A new well was drilled based on the subsidence data, a reinterpretation of the structural model, and a revised flow model. Almost virgin pressure was found. The well has since been the best producer on the field and paid back the cost of data acquisition by at least two orders of magnitude.

View Video Presentation: https://doi.org/10.2514/6.2022-1172.vid An approach for dynamically adapting body-fitted overset Strand meshes in HPCMP CREATETM-AV Helios is demonstrated. The approach couples the Strand-Cartesian near-body/off-body meshing paradigm with the HPCMP CREATETM-FT Capstone software to generate and refine surface meshes. Helios then constructs the dual volume mesh and interpolates data to the adapted mesh. The approach is parallel and fully automated, requiring no intervention on the part of the user during the adaptation process. Tests are conducted for RANS calculations on the ONERA M6 wing configuration, both static and pitching. With the pitching wing, the adaptive process completes 192 time-dependent adapt cycles, maintaining good quality and good solution accuracy. On 440 processors, the adaptive process incurred less than 8% computation overhead.

View Video Presentation: https://doi.org/10.2514/6.2022-1175.vid Overset methods, when utilized within a multi-solver framework, are a very flexible and powerful toolset for analyzing moving body problems. The Parallel Unsteady Do- main Information Technology (PUNDIT) module is the dedicated overset grid assembly module that is used within both HPCMP CREATETM-AV Helios and Kestrel. PUNDIT is fully parallel and uses the automated implicit hole-cutting technique. Several development efforts are being pursued such as (1) Improved representation of body hole maps (2) Extension of domain connectivity algorithms to be amenable for GPU computing (3) investigation of high-order conservative interpolation algorithms and (4) General grid-to-grid interpolation techniques in the context of including atmospheric boundary layers. The methodologies and results obtained in these research areas are described in this manuscript.

We first exhibit a multimodal image registration task, for which a neural network trained on a dataset with noisy labels reaches almost perfect accuracy, far beyond noise variance. This surprising auto-denoising phenomenon can be explained as a noise averaging effect over the labels of similar input examples. This effect theoretically grows with the number of similar examples; the question is then to define and estimate the similarity of examples. We express a proper definition of similarity, from the neural network perspective, i.e. we quantify how undissociable two inputs $A$ and $B$ are, taking a machine learning viewpoint: how much a parameter variation designed to change the output for $A$ would impact the output for $B$ as well? We study the mathematical properties of this similarity measure, and show how to use it on a trained network to estimate sample density, in low complexity, enabling new types of statistical analysis for neural networks. We analyze data by retrieving samples perceived as similar by the network, and are able to quantify the denoising effect without requiring true labels. We also propose, during training, to enforce that examples known to be similar should also be seen as similar by the network, and notice speed-up training effects for certain datasets.

The HPCMP CREATE[TM]-AV Helios overset CFD framework has for long used patch-based Cartesian solver SAMCart for adaptive mesh refinement in off-body meshes. This work presents the development of a new solver, Orchard, designed to improve upon three primary aspects of the current approach. First, Orchard uses structured sets of linear octrees to improve scalability and enable growth to exa-scale simulations. Second, Orchard sub-divides each octant into fixed-dimension Cartesian blocks for cache-efficient algorithms at the block-level and global GMRES algorithm over all blocks. Finally, the framework is designed to run on either CPUs or GPUs to accelerate computation. Vortex convection-diffusion, forward-flight PSP rotor, and forward-flight HART-II rotor cases are presented to analyze accuracy and performance of the new Cartesian solver. On CPUs, Orchard out-performs SAMCart timing for each case. Orchard on GPU-accelerated nodes runs 12.8X faster than Orchard on typical CPU cluster nodes.