A tight inclusion function for continuous collision detection

Last update: Jan 01, 2023

Related tags

Overview

Tight-Inclusion Continuous Collision Detection

A conservative Continuous Collision Detection (CCD) method with support for minimum separation.

You can read more about this work in our ACM Transactions on Graphics paper:

"A Large Scale Benchmark and an Inclusion-Based Algorithm forContinuous Collision Detection"

@article{Wang:2021:Benchmark,
    title   = {A Large Scale Benchmark and an Inclusion-Based Algorithm for Continuous Collision Detection},
    author  = {Bolun Wang and Zachary Ferguson and Teseo Schneider and Xin Jiang and Marco Attene and Daniele Panozzo},
    year    = 2021,
    journal = {ACM Transactions on Graphics}
}

Compiling Instruction

To compile the code, first make sure CMake is installed.

To build the library on Linux or macOS:

mkdir build
cd build
cmake ../ -DCMAKE_BUILD_TYPE=Release
make

Then you can run a CCD example:

./Tight_Inclusion_bin

We also provide you an example to run the Sample Queries using our CCD method. You may need to install gmp before compiling the code. Then set the CMake option TIGHT_INCLUSION_WITH_TESTS as ON when compiling:

cmake ../ -DCMAKE_BUILD_TYPE=Release -DTIGHT_INCLUSION_WITH_TESTS=ON
make

Then you can run ./Tight_Inclusion_bin to test the handcrafted queries in the Sample Queries.

Usage

Include #include <tight_inclusion/inclusion_ccd.hpp>

To check edge-edge ccd, use bool inclusion_ccd::edgeEdgeCCD_double();

To check vertex-face ccd, use bool inclusion_ccd::vertexFaceCCD_double();

💡 If collision is detected, the ccd function will return true, otherwise, the ccd function will return false. Since our method is CONSERVATIVE, if the returned result is false, we guarantee that there is no collision happens. If the result is true, it is possible that there is no collision but we falsely report a collision, but we can guarantee that this happens only if the minimal distance between the two primitives in this time step is no larger than tolerance + ms + err. We wil explain these parameters below.

For both edge-edge ccd and vertex-face ccd, the input CCD query is presented by 8 vertices which are in the format of Eigen::Vector3d. Please read our code in tight_inclusion/inclusion_ccd.hpp for the correct input order of the vertices.

Beside the input vertices, there are some input and output parameters for users to tune the performace or to get more CCD information. Here is a list of the explanations of the parameters:

input:
    err                 The numerical filters of the x, y and z coordinates. It measures the errors introduced by floating-point calculation when solving inclusion functions.
    ms                  Minimum separation distance no less than 0. It guarantees that collision will be reported if the distance between the two primitives is less than ms.
    tolerance           User-specific solving precision. It is the target maximal x, y, and z length of the inclusion function. We suggest the user to set it as 1e-6.
    t_max               The time range [0, t_max] where we detect collisions. Since the input query implies the motion in time range [0, 1], t_max should no larger than 1.
    max_itr             The parameter to enable early termination of the algorithm. If you set max_itr < 0, early termination will be disabled, but this may cause longer runtime. We suggest to set max_itr = 1e6.
    CCD_TYPE            The parameter to choose collision detection algorithms. By default CCD_TYPE = 1. If set CCD_TYPE = 0, the code will switch to a naive conservative CCD algorithm, but lack of our advanced features. 
    
output:
    toi                 Time of impact. If multiple collisions happen in this time step, it will return the earlist collision time. If there is no collision, the returned toi value will be std::numeric_limits<double>::infinity().
    output_tolerance    The real solving precision. If early termination is enabled, the solving precision may not reach the target precision. This parameter will return the real solving precision when the code is terminated.

Tips

💡 The input parameter err is crucial to guarantee our algorithm to be a conservative method not affected by floating point rounding errors. To run a single query, you can set err = {{-1, -1, -1}} to enable a sub-function to calculate the real numerical filters when solving CCD. If you are integrating our CCD in simulators, you need to:

Include the headler: #include <tight_inclusion/interval_root_finder.hpp>.
Call std::array<double, 3> err_vf = inclusion_ccd::get_numerical_error() and std::array<double, 3> err_ee = inclusion_ccd::get_numerical_error()
Use the parameter err_ee each time you call bool inclusion_ccd::edgeEdgeCCD_double() and err_vf when you call bool inclusion_ccd::vertexFaceCCD_double().

The parameters for function inclusion_ccd::get_numerical_error() is explained below:

input:
    vertices            Vertices of the Axies-Aligned-Bounding-Box of the simulation scene. Before you run the simulation, you need to conservatively estimate the Axies-Aligned-Bounding-Box in which the meshes will located during the whole simulation process, and the vertices should be the corners of the AABB.
    check_vf            A boolean instruction showing if you are checking vertex-face or edge-edge CCD.
    using_minimum_separation    A boolean instruction. If you are using minimum-separation CCD (the input parameter ms > 0), please set it as true.

💡 For some simulators which use non-zero minimum separation distance (ms > 0) to make sure intersection-free for each time-step, we have a CMake option TIGHT_INCLUSION_WITH_NO_ZERO_TOI to avoid the returned collision time toi is 0. You need to set TIGHT_INCLUSION_WITH_NO_ZERO_TOI as ON when compiling by: cmake ../ -DCMAKE_BUILD_TYPE=Release -DTIGHT_INCLUSION_WITH_NO_ZERO_TOI=ON. Then when you use the CCD functions, the code will continue the refinement in higher precision if the output toi is 0 under the given tolerance. So, the eventually toi will not be 0.

To have a better understand, or to get more details of our Tight-Inclusion CCD algorithm, please refer to our paper.

Continuous Query Decomposition for Complex Query Answering in Incomplete Knowledge Graphs

Continuous Query Decomposition This repository contains the official implementation for our ICLR 2021 (Oral) paper, Complex Query Answering with Neura

71 Dec 29, 2022

On the model-based stochastic value gradient for continuous reinforcement learning

On the model-based stochastic value gradient for continuous reinforcement learning This repository is by Brandon Amos, Samuel Stanton, Denis Yarats, a

46 Dec 15, 2022

Continuous Diffusion Graph Neural Network

We present Graph Neural Diffusion (GRAND) that approaches deep learning on graphs as a continuous diffusion process and treats Graph Neural Networks (GNNs) as discretisations of an underlying PDE.

227 Jan 5, 2023

PyTorch implementation for MINE: Continuous-Depth MPI with Neural Radiance Fields

MINE: Continuous-Depth MPI with Neural Radiance Fields Project Page | Video PyTorch implementation for our ICCV 2021 paper. MINE: Towards Continuous D

325 Dec 29, 2022

This repository contains the source code and data for reproducing results of Deep Continuous Clustering paper

Deep Continuous Clustering Introduction This is a Pytorch implementation of the DCC algorithms presented in the following paper (paper): Sohil Atul Sh

197 Nov 29, 2022

This is the codebase for the ICLR 2021 paper Trajectory Prediction using Equivariant Continuous Convolution

Trajectory Prediction using Equivariant Continuous Convolution (ECCO) This is the codebase for the ICLR 2021 paper Trajectory Prediction using Equivar

45 Jul 22, 2022

PyTorch implementation for Stochastic Fine-grained Labeling of Multi-state Sign Glosses for Continuous Sign Language Recognition.

Stochastic CSLR This is the PyTorch implementation for the ECCV 2020 paper: Stochastic Fine-grained Labeling of Multi-state Sign Glosses for Continuou

28 Dec 19, 2022

Code for the ICASSP-2021 paper: Continuous Speech Separation with Conformer.

Continuous Speech Separation with Conformer Introduction We examine the use of the Conformer architecture for continuous speech separation. Conformer

81 Nov 28, 2022

A clean and robust Pytorch implementation of PPO on continuous action space.

PPO-Continuous-Pytorch I found the current implementation of PPO on continuous action space is whether somewhat complicated or not stable. And this is

56 Dec 16, 2022

Comments

Improved No Zero ToI
I changed the options for avoiding a zero time of impact by making the option a parameter to the CCD functions.

I also changed from a recursive algorithm to an iterative one with a maximum number of iterations.
opened by zfergus 2

a case returned 0 TOI

For the PT CCD case below, TICCD returned 0 TOI.

3.5000000000e-01 0.0000000000e+00 3.5000000000e-01
3.4604643638e-01 2.7847887896e-03 3.4658121160e-01
3.5000819263e-01 -1.5763377304e-06 3.5001214242e-01
3.4903882032e-01 -1.5866575093e-03 3.4560164356e-01
0.0000000000e+00 0.0000000000e+00 0.0000000000e+00
1.0120300789e-07 -1.4009994248e-04 -8.5653091896e-05
-9.2976239634e-07 1.4361165221e-06 5.5168830145e-07
1.1649271683e-04 -3.6172565398e-05 -4.7128237917e-05

The format is

node_x node_y node_z
traingle_node0_x traingle_node0_y traingle_node0_z
traingle_node1_x traingle_node1_y traingle_node1_z
traingle_node2_x traingle_node2_y traingle_node2_z
node_displacement_x node_displacement_y node_displacement_z
traingle_node0_displacement_x traingle_node0_displacement_y traingle_node0_displacement_z
traingle_node1_displacement_x traingle_node1_displacement_y traingle_node1_displacement_z
traingle_node2_displacement_x traingle_node2_displacement_y traingle_node2_displacement_z

I used the TICCD from the CCDWrapper repo (latest commit), and I called it like

template <class T>
bool Point_Triangle_CCD_TI(
    const Eigen::Matrix<T, 3, 1>& p, 
    const Eigen::Matrix<T, 3, 1>& t0, 
    const Eigen::Matrix<T, 3, 1>& t1,
    const Eigen::Matrix<T, 3, 1>& t2,
    const Eigen::Matrix<T, 3, 1>& dp, 
    const Eigen::Matrix<T, 3, 1>& dt0, 
    const Eigen::Matrix<T, 3, 1>& dt1,
    const Eigen::Matrix<T, 3, 1>& dt2,
    T eta, T thickness, T& toc)
{
    T toc_prev = toc;
    T output_tolerance;

    T dist2_cur;
    Point_Triangle_Distance_Unclassified(p, t0, t1, t2, dist2_cur);
    T dist_cur = std::sqrt(dist2_cur);
    if (inclusion_ccd::vertexFaceCCD_double(p, t0, t1, t2,
        p + dp, t0 + dt0, t1 + dt1, t2 + dt2, { { -1, 0, 0 } },
        eta * (dist2_cur - thickness * thickness) / (dist_cur + thickness) + thickness,
        toc, 1e-6, toc_prev, 1e6, output_tolerance, 1))
    {
        if (toc < 1.0e-6) {
            std::cout << "PT CCD tiny!" << std::endl;
            if (inclusion_ccd::vertexFaceCCD_double(p, t0, t1, t2,
                p + dp, t0 + dt0, t1 + dt1, t2 + dt2, { { -1, 0, 0 } },
                thickness, toc, 1e-6, toc_prev, 1e6, output_tolerance, 1))
            {
                toc *= (1.0 - eta);
                return true;
            }
            else {
                return false;
            }
        }
        return true;
    }
    return false;
}

printf("TICCD:\n");
largestAlpha = 1;
tstart = clock();
if (Point_Triangle_CCD_TI(x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7], 0.1, 0.0, largestAlpha)) {
    printf("%le\n", largestAlpha);
}
else {
    printf("no collision\n");
}
printf("%.1lems\n", double(clock() - tstart) / CLOCKS_PER_SEC * 1000);

The output is

TICCD:
PT CCD tiny!
0.000000e+00
1.9e+00ms

However I implemented a simpler version of TICCD according to the paper as

template <class T>
void Point_Triangle_Distance_Vector_Unclassified(
    const Eigen::Matrix<T, 3, 1>& p, 
    const Eigen::Matrix<T, 3, 1>& t0, 
    const Eigen::Matrix<T, 3, 1>& t1,
    const Eigen::Matrix<T, 3, 1>& t2,
    const Eigen::Matrix<T, 3, 1>& dp, 
    const Eigen::Matrix<T, 3, 1>& dt0, 
    const Eigen::Matrix<T, 3, 1>& dt1,
    const Eigen::Matrix<T, 3, 1>& dt2,
    T t, T lambda, T beta,
    Eigen::Matrix<T, 3, 1>& distVec)
{
    const Eigen::Matrix<T, 3, 1> tp = (1 - lambda - beta) * t0 + lambda * t1 + beta * t2;
    const Eigen::Matrix<T, 3, 1> dtp = (1 - lambda - beta) * dt0 + lambda * dt1 + beta * dt2;
    distVec = p + t * dp - (tp + t * dtp);
}

template <class T>
bool Point_Triangle_CheckInterval_Unclassified(
    const Eigen::Matrix<T, 3, 1>& p, 
    const Eigen::Matrix<T, 3, 1>& t0, 
    const Eigen::Matrix<T, 3, 1>& t1,
    const Eigen::Matrix<T, 3, 1>& t2,
    const Eigen::Matrix<T, 3, 1>& dp, 
    const Eigen::Matrix<T, 3, 1>& dt0, 
    const Eigen::Matrix<T, 3, 1>& dt1,
    const Eigen::Matrix<T, 3, 1>& dt2,
    const std::array<T, 6>& interval,
    T gap)
{
    Eigen::Matrix<T, 3, 1> distVecMax, distVecMin;
    distVecMax.setConstant(-2 * gap - 1);
    distVecMin.setConstant(2 * gap + 1);
    for (int t = 0; t < 2; ++t) {
        for (int lambda = 0; lambda < 2; ++lambda) {
            for (int beta = 0; beta < 2; ++beta) {
                if (lambda == 1 && beta == 1) {
                    continue;
                }
                Eigen::Matrix<T, 3, 1> distVec;
                Point_Triangle_Distance_Vector_Unclassified(p, t0, t1, t2, dp, dt0, dt1, dt2, 
                    interval[t], interval[2 + lambda], interval[4 + beta], distVec);
                distVecMax = distVecMax.array().max(distVec.array());
                distVecMin = distVecMin.array().min(distVec.array());
            }
        }
    }
    return (distVecMax.array() >= -gap).all() && (distVecMin.array() <= gap).all();
}
template <class T>
bool Point_Triangle_TICCD(
    const Eigen::Matrix<T, 3, 1>& p, 
    const Eigen::Matrix<T, 3, 1>& t0, 
    const Eigen::Matrix<T, 3, 1>& t1,
    const Eigen::Matrix<T, 3, 1>& t2,
    Eigen::Matrix<T, 3, 1> dp, 
    Eigen::Matrix<T, 3, 1> dt0, 
    Eigen::Matrix<T, 3, 1> dt1,
    Eigen::Matrix<T, 3, 1> dt2,
    T eta, T thickness, T& toc)
{
    T dist2_cur;
    Point_Triangle_Distance_Unclassified(p, t0, t1, t2, dist2_cur);
    T dist_cur = std::sqrt(dist2_cur);
    T gap = eta * (dist2_cur - thickness * thickness) / (dist_cur + thickness);

    T tTol = 1e-3;

    std::vector<std::array<T, 6>> roots;
    std::deque<std::array<T, 6>> intervals;
    intervals.push_back({0, toc, 0, 1, 0, 1});
    int iterAmt = 0;
    while (!intervals.empty()) {
        ++iterAmt;

        std::array<T, 6> curIV = intervals.front();
        intervals.pop_front();

        // check
        if (Point_Triangle_CheckInterval_Unclassified(p, t0, t1, t2, dp, dt0, dt1, dt2, curIV, gap)) {
            if (curIV[0] && curIV[1] - curIV[0] < tTol) {
                // root found within tTol
                roots.emplace_back(curIV);
            }
            else {
                // split interval and push back
                std::vector<T> intervalLen({curIV[1] - curIV[0], curIV[3] - curIV[2], curIV[5] - curIV[4]});
                switch (std::max_element(intervalLen.begin(), intervalLen.end()) - intervalLen.begin()) {
                case 0:
                    intervals.push_back({curIV[0], (curIV[1] + curIV[0]) / 2, curIV[2], curIV[3], curIV[4], curIV[5]});
                    intervals.push_back({(curIV[1] + curIV[0]) / 2, curIV[1], curIV[2], curIV[3], curIV[4], curIV[5]});
                    break;

                case 1:
                    intervals.push_back({curIV[0], curIV[1], curIV[2], (curIV[2] + curIV[3]) / 2, curIV[4], curIV[5]});
                    intervals.push_back({curIV[0], curIV[1], (curIV[2] + curIV[3]) / 2, curIV[3], curIV[4], curIV[5]});
                    break;

                case 2:
                    intervals.push_back({curIV[0], curIV[1], curIV[2], curIV[3], curIV[4], (curIV[4] + curIV[5]) / 2});
                    intervals.push_back({curIV[0], curIV[1], curIV[2], curIV[3], (curIV[4] + curIV[5]) / 2, curIV[5]});
                    break;
                }
            }
        }
    }
    
    if (roots.empty()) {
        printf("TICCD PT converged with %d iters\n", iterAmt);
        return false;
    }
    else {
        for (const auto& rI : roots) {
            if (toc > rI[0]) {
                toc = rI[0];
            }
        }
        printf("TICCD PT converged with %d iters\n", iterAmt);
        return true;
    }
}

and it kinds of worked on this case and output

TICCD PT converged with 6143 iters
4.882812e-04
5.3e-01ms

which with tighter convergence tolerance might be able to reach the ground truth solution -- no collision.

In my implementation, I didn't include the epsilons in formula (5) in the paper. Is it the main cause here?

opened by liminchen 2

Refactored and cleaned up code

I tested and timed it on the entire dataset. I get 0 FN, a couple of fewer FP on the handcrafted data, and the timing is a little faster on the simulation and faster on handcrafted.

opened by zfergus 0

Releases(v1.0.2)

v1.0.2(Jun 30, 2022)

Resolved compiler warnings on Linux and macOS.
Source code(tar.gz)
Source code(zip)
v1.0.1(Jun 9, 2022)

Refactor with slight performance improvement
Source code(tar.gz)
Source code(zip)
v1.0.0(Jan 28, 2022)

This is a stable release of our CCD algorithm before we refactor some of the code and make updates to the API.
Source code(tar.gz)
Source code(zip)

Owner

Continuous Collision Detection

Code for continuous collision detection methods bechmarked in "A Large Scale Benchmark and an Inclusion-Based Algorithm for Continuous Collision Detection"

GitHub Repository

AWS documentation corpus for zero-shot open-book question answering.

aws-documentation We present the AWS documentation corpus, an open-book QA dataset, which contains 25,175 documents along with 100 matched questions a

2 Jul 07, 2022

CityLearn Challenge Multi-Agent Reinforcement Learning for Intelligent Energy Management, 2020, PikaPika team

Citylearn Challenge This is the PyTorch implementation for PikaPika team, CityLearn Challenge Multi-Agent Reinforcement Learning for Intelligent Energ

10 Oct 10, 2022

This repository lets you interact with Lean through a REPL.

lean-gym This repository lets you interact with Lean through a REPL. See Formal Mathematics Statement Curriculum Learning for a presentation of lean-g

87 Dec 28, 2022

Pytorch Implementation of the paper "Cross-domain Correspondence Learning for Exemplar-based Image Translation"

CoCosNet Pytorch Implementation of the paper "Cross-domain Correspondence Learning for Exemplar-based Image Translation" (CVPR 2020 oral). Update: 202

38 Sep 22, 2021

Fast Neural Style for Image Style Transform by Pytorch

FastNeuralStyle by Pytorch Fast Neural Style for Image Style Transform by Pytorch This is famous Fast Neural Style of Paper Perceptual Losses for Real

81 Sep 03, 2022

The code of Zero-shot learning for low-light image enhancement based on dual iteration

Zero-shot-dual-iter-LLE The code of Zero-shot learning for low-light image enhancement based on dual iteration. You can get the real night image tests

1 Mar 18, 2022

Deep RGB-D Saliency Detection with Depth-Sensitive Attention and Automatic Multi-Modal Fusion (CVPR'2021, Oral)

DSA^2 F: Deep RGB-D Saliency Detection with Depth-Sensitive Attention and Automatic Multi-Modal Fusion (CVPR'2021, Oral) This repo is the official imp

46 Dec 21, 2022

System Design course at HSE (2021)

System Design course at HSE (2021) Wiki-страница курса Структура репозитория: slides - директория с презентациями с занятий tasks - материалы для выпо

22 Dec 25, 2022

MultiMix: Sparingly Supervised, Extreme Multitask Learning From Medical Images (ISBI 2021, MELBA 2021)

MultiMix This repository contains the implementation of MultiMix. Our publications for this project are listed below: "MultiMix: Sparingly Supervised,

27 Dec 22, 2022

Deep and online learning with spiking neural networks in Python

Introduction The brain is the perfect place to look for inspiration to develop more efficient neural networks. One of the main differences with modern

447 Jan 03, 2023

Python package for missing-data imputation with deep learning

MIDASpy Overview MIDASpy is a Python package for multiply imputing missing data using deep learning methods. The MIDASpy algorithm offers significant

77 Dec 03, 2022

The code for the NeurIPS 2021 paper "A Unified View of cGANs with and without Classifiers".

Energy-based Conditional Generative Adversarial Network (ECGAN) This is the code for the NeurIPS 2021 paper "A Unified View of cGANs with and without

22 May 28, 2022

For auto aligning, cropping, and scaling HR and LR images for training image based neural networks

ImgAlign For auto aligning, cropping, and scaling HR and LR images for training image based neural networks Usage Make sure OpenCV is installed, 'pip

15 Dec 04, 2022

AntroPy: entropy and complexity of (EEG) time-series in Python

AntroPy is a Python 3 package providing several time-efficient algorithms for computing the complexity of time-series. It can be used for example to e

153 Dec 27, 2022

EMNLP 2021 - Frustratingly Simple Pretraining Alternatives to Masked Language Modeling

Frustratingly Simple Pretraining Alternatives to Masked Language Modeling This is the official implementation for "Frustratingly Simple Pretraining Al

31 Nov 18, 2022

10x faster matrix and vector operations

Bolt is an algorithm for compressing vectors of real-valued data and running mathematical operations directly on the compressed representations. If yo

2.3k Jan 09, 2023

The repository contains reproducible PyTorch source code of our paper Generative Modeling with Optimal Transport Maps, ICLR 2022.

Generative Modeling with Optimal Transport Maps The repository contains reproducible PyTorch source code of our paper Generative Modeling with Optimal

30 Dec 22, 2022