COccupancyGridMap2D_simulate.cpp

Go to the documentation of this file.

/* +------------------------------------------------------------------------+
   |                     Mobile Robot Programming Toolkit (MRPT)            |
   |                          https://www.mrpt.org/                         |
   |                                                                        |
   | Copyright (c) 2005-2019, Individual contributors, see AUTHORS file     |
   | See: https://www.mrpt.org/Authors - All rights reserved.               |
   | Released under BSD License. See: https://www.mrpt.org/License          |
   +------------------------------------------------------------------------+ */

#include "maps-precomp.h"  // Precomp header

#include <mrpt/core/round.h>  // round()
#include <mrpt/maps/COccupancyGridMap2D.h>
#include <mrpt/math/CVectorFixed.h>
#include <mrpt/math/transform_gaussian.h>
#include <mrpt/obs/CObservation2DRangeScan.h>
#include <mrpt/obs/CObservationRange.h>
#include <mrpt/random.h>
#include <Eigen/Dense>

using namespace mrpt;
using namespace mrpt::maps;
using namespace mrpt::obs;
using namespace mrpt::random;
using namespace mrpt::poses;
using namespace std;

double COccupancyGridMap2D::RAYTRACE_STEP_SIZE_IN_CELL_UNITS = 0.8;

// See docs in header
void COccupancyGridMap2D::laserScanSimulator(
    mrpt::obs::CObservation2DRangeScan& inout_Scan, const CPose2D& robotPose,
    float threshold, size_t N, float noiseStd, unsigned int decimation,
    float angleNoiseStd) const
{
    MRPT_START

    ASSERT_(decimation >= 1);
    ASSERT_(N >= 2);

    // Sensor pose in global coordinates
    CPose3D sensorPose3D = CPose3D(robotPose) + inout_Scan.sensorPose;
    // Aproximation: grid is 2D !!!
    CPose2D sensorPose(sensorPose3D);

    // Scan size:
    inout_Scan.resizeScan(N);

    double A = sensorPose.phi() +
               (inout_Scan.rightToLeft ? -0.5 : +0.5) * inout_Scan.aperture;
    const double AA =
        (inout_Scan.rightToLeft ? 1.0 : -1.0) * (inout_Scan.aperture / (N - 1));

    const float free_thres = 1.0f - threshold;

    for (size_t i = 0; i < N; i += decimation, A += AA * decimation)
    {
        bool valid;
        float out_range;
        simulateScanRay(
            sensorPose.x(), sensorPose.y(), A, out_range, valid,
            inout_Scan.maxRange, free_thres, noiseStd, angleNoiseStd);
        inout_Scan.setScanRange(i, out_range);
        inout_Scan.setScanRangeValidity(i, valid);
    }

    MRPT_END
}

void COccupancyGridMap2D::sonarSimulator(
    CObservationRange& inout_observation, const CPose2D& robotPose,
    float threshold, float rangeNoiseStd, float angleNoiseStd) const
{
    const float free_thres = 1.0f - threshold;

    for (auto itR = inout_observation.begin(); itR != inout_observation.end();
         ++itR)
    {
        const CPose2D sensorAbsolutePose =
            CPose2D(CPose3D(robotPose) + CPose3D(itR->sensorPose));

        // For each sonar cone, simulate several rays and keep the shortest
        // distance:
        ASSERT_(inout_observation.sensorConeApperture > 0);
        size_t nRays =
            round(1 + inout_observation.sensorConeApperture / DEG2RAD(1.0));

        double direction = sensorAbsolutePose.phi() -
                           0.5 * inout_observation.sensorConeApperture;
        const double Adir = inout_observation.sensorConeApperture / nRays;

        float min_detected_obs = 0;
        for (size_t i = 0; i < nRays; i++, direction += Adir)
        {
            bool valid;
            float sim_rang;
            simulateScanRay(
                sensorAbsolutePose.x(), sensorAbsolutePose.y(), direction,
                sim_rang, valid, inout_observation.maxSensorDistance,
                free_thres, rangeNoiseStd, angleNoiseStd);

            if (valid && (sim_rang < min_detected_obs || !i))
                min_detected_obs = sim_rang;
        }
        // Save:
        itR->sensedDistance = min_detected_obs;
    }
}

void COccupancyGridMap2D::simulateScanRay(
    const double start_x, const double start_y, const double angle_direction,
    float& out_range, bool& out_valid, const double max_range_meters,
    const float threshold_free, const double noiseStd,
    const double angleNoiseStd) const
{
    const double A_ =
        angle_direction +
        (angleNoiseStd > .0
             ? getRandomGenerator().drawGaussian1D_normalized() * angleNoiseStd
             : .0);

// Unit vector in the directorion of the ray:
#ifdef HAVE_SINCOS
    double Arx, Ary;
    ::sincos(A_, &Ary, &Arx);
#else
    const double Arx = cos(A_);
    const double Ary = sin(A_);
#endif

    // Ray tracing, until collision, out of the map or out of range:
    const unsigned int max_ray_len = mrpt::round(max_range_meters / resolution);
    unsigned int ray_len = 0;

// Use integers for all ray tracing for efficiency
#define INTPRECNUMBIT 10
#define int_x2idx(_X) (_X >> INTPRECNUMBIT)
#define int_y2idx(_Y) (_Y >> INTPRECNUMBIT)

    auto rxi = static_cast<int64_t>(
        ((start_x - x_min) / resolution) * (1L << INTPRECNUMBIT));
    auto ryi = static_cast<int64_t>(
        ((start_y - y_min) / resolution) * (1L << INTPRECNUMBIT));

    const auto Arxi = static_cast<int64_t>(
        RAYTRACE_STEP_SIZE_IN_CELL_UNITS * Arx * (1L << INTPRECNUMBIT));
    const auto Aryi = static_cast<int64_t>(
        RAYTRACE_STEP_SIZE_IN_CELL_UNITS * Ary * (1L << INTPRECNUMBIT));

    cellType hitCellOcc_int = 0;  // p2l(0.5f)
    const cellType threshold_free_int = p2l(threshold_free);
    int x, y = int_y2idx(ryi);

    while ((x = int_x2idx(rxi)) >= 0 && (y = int_y2idx(ryi)) >= 0 &&
           x < static_cast<int>(size_x) && y < static_cast<int>(size_y) &&
           (hitCellOcc_int = map[x + y * size_x]) > threshold_free_int &&
           ray_len < max_ray_len)
    {
        rxi += Arxi;
        ryi += Aryi;
        ray_len++;
    }

    // Store:
    // Check out of the grid?
    // Tip: if x<0, (unsigned)(x) will also be >>> size_x ;-)
    if (abs(hitCellOcc_int) <= 1 || static_cast<unsigned>(x) >= size_x ||
        static_cast<unsigned>(y) >= size_y)
    {
        out_valid = false;
        out_range = max_range_meters;
    }
    else
    {  // No: The normal case:
        out_range = RAYTRACE_STEP_SIZE_IN_CELL_UNITS * ray_len * resolution;
        out_valid = (ray_len < max_ray_len);  // out_range<max_range_meters;
        // Add additive Gaussian noise:
        if (noiseStd > 0 && out_valid)
            out_range +=
                noiseStd * getRandomGenerator().drawGaussian1D_normalized();
    }
}

COccupancyGridMap2D::TLaserSimulUncertaintyParams::
	TLaserSimulUncertaintyParams()

    = default;

COccupancyGridMap2D::TLaserSimulUncertaintyResult::
	TLaserSimulUncertaintyResult() = default;

struct TFunctorLaserSimulData
{
    const COccupancyGridMap2D::TLaserSimulUncertaintyParams* params;
    const COccupancyGridMap2D* grid;
};

static void func_laserSimul_callback(
    const mrpt::math::CVectorFixedDouble<3>& x_pose,
    const TFunctorLaserSimulData& fixed_param,
    mrpt::math::CVectorDouble& y_scanRanges)
{
    ASSERT_(fixed_param.params && fixed_param.grid);
    ASSERT_(fixed_param.params->decimation >= 1);
    ASSERT_(fixed_param.params->nRays >= 2);

    const size_t N = fixed_param.params->nRays;

    // Sensor pose in global coordinates
    const CPose3D sensorPose3D =
        CPose3D(x_pose[0], x_pose[1], .0, x_pose[2], .0, .0) +
        fixed_param.params->sensorPose;
    // Aproximation: grid is 2D !!!
    const CPose2D sensorPose(sensorPose3D);

    // Scan size:
    y_scanRanges.resize(N);

    double A =
        sensorPose.phi() + (fixed_param.params->rightToLeft ? -0.5 : +0.5) *
                               fixed_param.params->aperture;
    const double AA = (fixed_param.params->rightToLeft ? 1.0 : -1.0) *
                      (fixed_param.params->aperture / (N - 1));

    const float free_thres = 1.0f - fixed_param.params->threshold;

    for (size_t i = 0; i < N; i += fixed_param.params->decimation,
                A += AA * fixed_param.params->decimation)
    {
        bool valid;
        float range;

        fixed_param.grid->simulateScanRay(
            sensorPose.x(), sensorPose.y(), A, range, valid,
            fixed_param.params->maxRange, free_thres, .0 /*noiseStd*/,
            .0 /*angleNoiseStd*/);
        y_scanRanges[i] = valid ? range : fixed_param.params->maxRange;
    }
}

void COccupancyGridMap2D::laserScanSimulatorWithUncertainty(
    const COccupancyGridMap2D::TLaserSimulUncertaintyParams& in_params,
    COccupancyGridMap2D::TLaserSimulUncertaintyResult& out_results) const
{
    const mrpt::math::CVectorFixedDouble<3> robPoseMean =
        in_params.robotPose.mean.asVectorVal();

    TFunctorLaserSimulData simulData;
    simulData.grid = this;
    simulData.params = &in_params;

    switch (in_params.method)
    {
        case sumUnscented:
            mrpt::math::transform_gaussian_unscented(
                robPoseMean,  // x_mean
                in_params.robotPose.cov,  // x_cov
                &func_laserSimul_callback,  // void  (*functor)(const
                // VECTORLIKE1 &x,const USERPARAM
                // &fixed_param, VECTORLIKE3 &y)
                simulData,  // const USERPARAM &fixed_param,
                out_results.scanWithUncert.rangesMean,
                out_results.scanWithUncert.rangesCovar,
                nullptr,  // elem_do_wrap2pi,
                in_params.UT_alpha, in_params.UT_kappa,
                in_params.UT_beta  // alpha, K, beta
            );
            break;
        case sumMonteCarlo:
            //
            mrpt::math::transform_gaussian_montecarlo(
                robPoseMean,  // x_mean
                in_params.robotPose.cov,  // x_cov
                &func_laserSimul_callback,  // void  (*functor)(const
                // VECTORLIKE1 &x,const USERPARAM
                // &fixed_param, VECTORLIKE3 &y)
                simulData,  // const USERPARAM &fixed_param,
                out_results.scanWithUncert.rangesMean,
                out_results.scanWithUncert.rangesCovar, in_params.MC_samples);
            break;
        default:
            throw std::runtime_error(
                "[laserScanSimulatorWithUncertainty] Unknown `method` value");
            break;
    };

    // Outputs:
    out_results.scanWithUncert.rangeScan.aperture = in_params.aperture;
    out_results.scanWithUncert.rangeScan.maxRange = in_params.maxRange;
    out_results.scanWithUncert.rangeScan.rightToLeft = in_params.rightToLeft;
    out_results.scanWithUncert.rangeScan.sensorPose = in_params.sensorPose;

    out_results.scanWithUncert.rangeScan.resizeScan(in_params.nRays);
    for (unsigned i = 0; i < in_params.nRays; i++)
    {
        out_results.scanWithUncert.rangeScan.setScanRange(
            i, (float)out_results.scanWithUncert.rangesMean[i]);
        out_results.scanWithUncert.rangeScan.setScanRangeValidity(i, true);
    }

    // Add minimum uncertainty: grid cell resolution:
    out_results.scanWithUncert.rangesCovar.asEigen().diagonal().array() +=
        0.5 * resolution * resolution;
}