Search

We introduce a novel differentiable hybrid traffic simulator, which simulates traffic using a hybrid model of both macroscopic and microscopic models and can be directly integrated into a neural network for traffic control and flow optimization. This is the first differentiable traffic simulator for macroscopic and hybrid models that can compute gradients for traffic states across time steps and inhomogeneous lanes. To compute the gradient flow between two types of traffic models in a hybrid framework, we present a novel intermediate conversion component that bridges the lanes in a differentiable manner as well. We also show that we can use analytical gradients to accelerate the overall process and enhance scalability. Thanks to these gradients, our simulator can provide more efficient and scalable solutions for complex learning and control problems posed in traffic engineering than other existing algorithms. Refer to https://sites.google.com/umd.edu/diff-hybrid-traffic-sim for our project., 13 pages, Siggraph Asia 2022 Journal Paper

In physically-based simulation, it is essential to choose appropriate material parameters to generate desirable simulation results. In many cases, however, choosing appropriate material parameters is very challenging, and often tedious trial-and-error parameter tuning steps are inevitable. In this paper, we propose a real-to-virtual parameter transfer framework that identifies material parameters of viscous fluids with example video data captured from real-world phenomena. Our method first extracts positional data of fluids and then uses the extracted data as a reference to identify the viscosity parameters, combining forward viscous fluid simulations and parameter optimization in an iterative process. We evaluate our method with a range of synthetic and real-world example data, and demonstrate that our method can identify the hidden physical variables and viscosity parameters. This set of recovered physical variables and parameters can then be effectively used in novel scenarios to generate viscous fluid behaviors visually consistent with the example videos.

Most recent garment capturing techniques rely on acquiring multiple views of clothing, which may not always be readily available, especially in the case of pre-existing photographs from the web. As an alternative, we propose a method that is able to compute a 3D model of a human body and its outfit from a single photograph with little human interaction. Our algorithm is not only able to capture the global shape and overall geometry of the clothing, it can also extract the physical properties (i.e., material parameters needed for simulation) of cloth. Unlike previous methods using full 3D information (i.e., depth, multi-view images, or sampled 3D geometry), our approach achieves garment recovery from a single-view image by using physical, statistical, and geometric priors and a combination of parameter estimation, semantic parsing, shape/pose recovery, and physics-based cloth simulation. We demonstrate the effectiveness of our algorithm by re-purposing the reconstructed garments for virtual try-on and garment transfer applications and for cloth animation on digital characters.

We present a novel hybrid approach that couples geometric and numerical acoustic techniques for interactive sound propagation in complex environments. Our formulation is based on a combination of spatial and frequency decomposition of the sound field. We use numerical wave-based techniques to precompute the pressure field in the near-object regions and geometric propagation techniques in the far-field regions to model sound propagation. We present a novel two-way pressure coupling technique at the interface of near-object and far-field regions. At runtime, the impulse response at the listener position is computed at interactive rates based on the stored pressure field and interpolation techniques. Our system is able to simulate high-fidelity acoustic effects such as diffraction, scattering, low-pass filtering behind obstruction, reverberation, and high-order reflections in large, complex indoor and outdoor environments and Half-Life 2 game engine. The pressure computation requires orders of magnitude lower memory than standard wave-based numerical techniques.

Linear modal synthesis methods have often been used to generate sounds for rigid bodies. One of the key challenges in widely adopting such techniques is the lack of automatic determination of satisfactory material parameters that recreate realistic audio quality of sounding materials. We introduce a novel method using prerecorded audio clips to estimate material parameters that capture the inherent quality of recorded sounding materials. Our method extracts perceptually salient features from audio examples. Based on psychoacoustic principles, we design a parameter estimation algorithm using an optimization framework and these salient features to guide the search of the best material parameters for modal synthesis. We also present a method that compensates for the differences between the real-world recording and sound synthesized using solely linear modal synthesis models to create the final synthesized audio. The resulting audio generated from this sound synthesis pipeline well preserves the same sense of material as a recorded audio example. Moreover, both the estimated material parameters and the residual compensation naturally transfer to virtual objects of different sizes and shapes, while the synthesized sounds vary accordingly. A perceptual study shows the results of this system compare well with real-world recordings in terms of material perception.

Simulating fluids in large-scale scenes with appreciable quality using state-of-the-art methods can lead to high memory and compute requirements. Since memory requirements are proportional to the product of domain dimensions, simulation performance is limited by memory access, as solvers for elliptic problems are not compute-bound on modern systems. This is a significant concern for large-scale scenes. To reduce the memory footprint and memory/compute ratio, vortex singularity bases can be used. Though they form a compact bases for incompressible vector fields, robust and efficient modeling of nonrigid obstacles and free-surfaces can be challenging with these methods. We propose a hybrid domain decomposition approach that couples Eulerian velocity-based simulations with vortex singularity simulations. Our formulation reduces memory footprint by using smaller Eulerian domains with compact vortex bases, thereby improving the memory/compute ratio, and simulation performance by more than 1000x for single phase flows as well as significant improvements for free-surface scenes. Coupling these two heterogeneous methods also affords flexibility in using the most appropriate method for modeling different scene features, as well as allowing robust interaction of vortex methods with free-surfaces and nonrigid obstacles.

We present an information-theoretic method to measure the similarity between a given set of observed, real-world data and visual simulation technique for aggregate crowd motions of a complex system consisting of many individual agents. This metric uses a two-step process to quantify a simulator's ability to reproduce the collective behaviors of the whole system, as observed in the recorded real-world data. First, Bayesian inference is used to estimate the simulation states which best correspond to the observed data, then a maximum likelihood estimator is used to approximate the prediction errors. This process is iterated using the EM-algorithm to produce a robust, statistical estimate of the magnitude of the prediction error as measured by its entropy (smaller is better). This metric serves as a simulator-to-data similarity measurement. We evaluated the metric in terms of robustness to sensor noise, consistency across different datasets and simulation methods, and correlation to perceptual metrics.

We present a method for real-time sound propagation that captures all wave effects, including diffraction and reverberation, for multiple moving sources and a moving listener in a complex, static 3D scene. It performs an offline numerical simulation over the scene and then applies a novel technique to extract and compactly encode the perceptually salient information in the resulting acoustic responses. Each response is automatically broken into two phases: early reflections (ER) and late reverberation (LR), via a threshold on the temporal density of arriving wavefronts. The LR is simulated and stored in the frequency domain, once per room in the scene. The ER accounts for more detailed spatial variation, by recording a set of peak delays/amplitudes in the time domain and a residual frequency response sampled in octave frequency bands, at each source/receiver point pair in a 5D grid. An efficient run-time uses this precomputed representation to perform binaural sound rendering based on frequency-domain convolution. Our system demonstrates realistic, wave-based acoustic effects in real time, including diffraction low-passing behind obstructions, sound focusing, hollow reverberation in empty rooms, sound diffusion in fully-furnished rooms, and realistic late reverberation.

We present a novel approach for synthesizing liquid sounds directly from visual simulation of fluid dynamics. Our approach takes advantage of the fact that the sound generated by liquid is mainly due to the vibration of resonating bubbles in the medium and performs automatic sound synthesis by coupling physically-based equations for bubble resonance with multiple fluid simulators. We effectively demonstrate our system on several benchmarks using a real-time shallow-water fluid simulator as well as a hybrid grid-SPH simulator.

Large dense crowds show aggregate behavior with reduced individual freedom of movement. We present a novel, scalable approach for simulating such crowds, using a dual representation both as discrete agents and as a single continuous system. In the continuous setting, we introduce a novel variational constraint called unilateral incompressibility , to model the large-scale behavior of the crowd, and accelerate inter-agent collision avoidance in dense scenarios. This approach makes it possible to simulate very large, dense crowds composed of up to a hundred thousand agents at near-interactive rates on desktop computers.

We present a novel technique for the animation of turbulent fluids by coupling a procedural turbulence model with a numerical fluid solver to introduce subgrid-scale flow detail. From the large-scale flow simulated by the solver, we model the production and behavior of turbulent energy using a physically motivated energy model. This energy distribution is used to synthesize an incompressible turbulent velocity field, whose features show plausible temporal behavior through a novel Lagrangian approach for advected noise. The synthesized turbulent flow has a dynamical effect on the large-scale flow, and produces visually plausible detailed features on both gaseous and free-surface liquid flows. Our method is an order of magnitude faster than full numerical simulation of equivalent resolution, and requires no manual direction.

We present a novel algorithm for accurately detecting all contacts, including self-collisions, between deformable models. We precompute a chromatic decomposition of a mesh into non-adjacent primitives using graph coloring algorithms. The chromatic decomposition enables us to check for collisions between non-adjacent primitives using a linear-time culling algorithm. As a result, we achieve higher culling efficiency and significantly reduce the number of false positives. We use our algorithm to check for collisions among complex deformable models consisting of tens of thousands of triangles for cloth modeling and medical simulation. Our algorithm accurately computes all contacts at interactive rates. We observed up to an order of magnitude speedup over prior methods.

We introduce a novel "sensation preserving" simplification algorithm for faster collision queries between two polyhedral objects in haptic rendering. Given a polyhedral model, we construct a multiresolution hierarchy using " filtered edge collapse", subject to constraints imposed by collision detection. The resulting hierarchy is then used to compute fast contact response for haptic display. The computation model is inspired by human tactual perception of contact information. We have successfully applied and demonstrated the algorithm on a time-critical collision query framework for haptically displaying complex object-object interaction. Compared to existing exact contact query algorithms, we observe noticeable performance improvement in update rates with little degradation in the haptic perception of contacts.