Search

Inspired by our previous work on mitigating the Kolmogorov barrier using a quadratic approximation manifold, we propose in this paper a computationally tractable approach for combining a projection-based reduced-order model (PROM) and an artificial neural network (ANN) for mitigating the Kolmogorov barrier to reducibility of convection-dominated flow problems. The main objective the PROM-ANN concept that we propose is to reduce the dimensionality of the online approximation of the solution beyond what is possible using affine and quadratic approximation manifolds. In contrast to previous approaches for constructing arbitrarily nonlinear manifold approximations for nonlinear model reduction that exploited one form or another of ANN, the training of the PROM-ANN we propose in this paper does not involve data whose dimension scales with that of the high-dimensional model; and this PROM-ANN is hyperreducible using any well-established hyperreduction method. Hence, unlike many other ANN-based approaches, the PROM-ANN concept we propose in this paper is practical for large-scale and industry-relevant CFD problems. Its potential is demonstrated here for a parametric, shock-dominated, benchmark problem.

We propose a method to engineer repulsive long-range interactions between ultracold ground-state molecules using optical fields, thus preventing short-range collisional losses. It maps the microwave coupling recently used for collisional shielding onto a two-photon transition, and takes advantage of optical control techniques. In contrast to one-photon optical shielding [Phys. Rev. Lett. 125, 153202 (2020)], this scheme avoids heating of the molecular gas due to photon scattering. The proposed protocol, exemplified for 23Na39K, should be applicable to a large class of polar diatomic molecules.

In this paper, we propose a new framework for multi-agent collaborative exploration of unknown environments. The proposed method combines state-of-the-art algorithms in mapping, safe corridor generation and multi-agent planning. It first takes a volume that we want to explore, then proceeds to give the multiple agents different goals in order to explore a voxel grid of that volume. The exploration ends when all voxels are discovered as free or occupied, or there is no path found for the remaining undiscovered voxels. The state-of-the-art planning algorithm uses time-aware Safe Corridors to guarantee intra-agent collision safety as well safety from static obstacles. The presented approach is tested in a state of the art simulator for up to 4 agents.

In this paper, we propose a new method for multirotor planning in dynamic environments. The environment is represented as a temporal occupancy grid which gives the current as well as the future/predicted state of all the obstacles. The method builds on previous works in Safe Corridor generation and multirotor planning to avoid moving and static obstacles. It first generates a global path to the goal that doesn't take into account the dynamic aspect of the environment. We then use temporal Safe Corridors to generate safe spaces that the robot can be in at discrete instants in the future. Finally we use the temporal Safe Corridors in an optimization formulation that accounts for the multirotor dynamics as well as all the obstacles to generate the trajectory that will be executed by the multirotor's controller. We show the performance of our method in simulations.

Safe Corridors (a series of overlapping convex shapes) have been used recently in multiple state-of-the-art motion planning methods. They allow to represent the free space in the environment in an efficient way for collision avoidance. In this paper, we propose a new framework for generating Safe Corridors. We assume that we have a voxel grid representation of the environment. The proposed framework improves on a previous state-of-the-art voxel grid based Safe Corridor generation method. It also creates a connectivity graph between polyhedra of a given Safe Corridor that allows to know which polyhedra intersect with each other. The connectivity graph can be used in planning methods to reduce computation time. The method is compared to other state-of-the-art methods in simulations in terms of computation time, volume covered, safety, number of polyhedron per Safe Corridor and number of constraints per polyhedron.

Training and testing supervised object detection models require a large collection of images with ground truth labels. Labels define object classes in the image, as well as their locations, shape, and possibly other information such as pose. The labeling process has proven extremely time consuming, even with the presence of manpower. We introduce a novel labeling tool for 2D images as well as 3D triangular meshes: 3D Labeling Tool (3DLT). This is a standalone, feature-heavy and cross-platform software that does not require installation and can run on Windows, macOS and Linux-based distributions. Instead of labeling the same object on every image separately like current tools, we use depth information to reconstruct a triangular mesh from said images and label the object only once on the aforementioned mesh. We use registration to simplify 3D labeling, outlier detection to improve 2D bounding box calculation and surface reconstruction to expand labeling possibility to large point clouds. Our tool is tested against state of the art methods and it greatly surpasses them in terms of speed while preserving accuracy and ease of use.

The hydrodynamic forces on a slender rod in a fluid medium at low Reynolds number can be modeled using resistive force theories (RFTs) or slender body theories (SBTs). The former represent the forces by local drag coefficients and are computationally cheap; however, they are physically inaccurate when long-range hydrodynamic interaction is involved. The later are physically accurate but require solving integral equations and, therefore, are computationally expensive. This paper investigates RFTs in comparison with state-of-the art SBT methods. During the process, a neural network-based hydrodynamic model that -- similar to RFTs -- relies on local drag coefficients for computational efficiency was developed. However, the network is trained using data from an SBT (regularized stokeslet segments method). The $R^2$ value of the trained coefficients were $\sim 0.99$ with mean absolute error of $1.6\times10^{-2}$. The machine learning resistive force theory (MLRFT) accounts for local hydrodynamic forces distribution, the dependence on rotational and translational speeds and directions, and geometric parameters of the slender object. We show that, when classical RFT fails to accurately predict the forces, torques, and drags on slender rods under low Reynolds number flows, MLRFT exhibits good agreement with physically accurate SBT simulations. In terms of computational speed, MLRFT forgoes the need of solving an inverse problem and, therefore, requires negligible computation time in comparison with SBT. MLRFT presents a computationally inexpensive hydrodynamic model for flagellar propulsion can be used in the design and optimization of biomimetic flagellated robots and analysis of bacterial locomotion., Comment: 9 pages, 10 figures

Data clipping is crucial in reducing noise in quantization operations and improving the achievable accuracy of quantization-aware training (QAT). Current practices rely on heuristics to set clipping threshold scalars and cannot be shown to be optimal. We propose Optimally Clipped Tensors And Vectors (OCTAV), a recursive algorithm to determine MSE-optimal clipping scalars. Derived from the fast Newton-Raphson method, OCTAV finds optimal clipping scalars on the fly, for every tensor, at every iteration of the QAT routine. Thus, the QAT algorithm is formulated with provably minimum quantization noise at each step. In addition, we reveal limitations in common gradient estimation techniques in QAT and propose magnitude-aware differentiation as a remedy to further improve accuracy. Experimentally, OCTAV-enabled QAT achieves state-of-the-art accuracy on multiple tasks. These include training-from-scratch and retraining ResNets and MobileNets on ImageNet, and Squad fine-tuning using BERT models, where OCTAV-enabled QAT consistently preserves accuracy at low precision (4-to-6-bits). Our results require no modifications to the baseline training recipe, except for the insertion of quantization operations where appropriate., Comment: Published as a spotlight paper at ICML 2022. Paper contains 16 pages, 5 figures, and 6 tables