| 1 | /** |
| 2 | * @file ISAM2_SmartFactorStereo_IMU.cpp |
| 3 | * @brief test of iSAM2 with smart stereo factors and IMU preintegration, |
| 4 | * originally used to debug valgrind invalid reads with Eigen |
| 5 | * @author Nghia Ho |
| 6 | * |
| 7 | * Setup is a stationary stereo camera with an IMU attached. |
| 8 | * The data file is at examples/Data/ISAM2_SmartFactorStereo_IMU.txt |
| 9 | * It contains 5 frames of stereo matches and IMU data. |
| 10 | */ |
| 11 | #include <gtsam/navigation/CombinedImuFactor.h> |
| 12 | #include <gtsam/nonlinear/ISAM2.h> |
| 13 | #include <gtsam_unstable/slam/SmartStereoProjectionPoseFactor.h> |
| 14 | |
| 15 | #include <fstream> |
| 16 | #include <iostream> |
| 17 | #include <sstream> |
| 18 | #include <string> |
| 19 | #include <vector> |
| 20 | |
| 21 | using namespace std; |
| 22 | using namespace gtsam; |
| 23 | using symbol_shorthand::B; |
| 24 | using symbol_shorthand::V; |
| 25 | using symbol_shorthand::X; |
| 26 | |
| 27 | struct IMUHelper { |
| 28 | IMUHelper() { |
| 29 | { |
| 30 | auto gaussian = noiseModel::Diagonal::Sigmas( |
| 31 | sigmas: (Vector(6) << Vector3::Constant(value: 5.0e-2), Vector3::Constant(value: 5.0e-3)) |
| 32 | .finished()); |
| 33 | auto huber = noiseModel::Robust::Create( |
| 34 | robust: noiseModel::mEstimator::Huber::Create(k: 1.345), noise: gaussian); |
| 35 | |
| 36 | biasNoiseModel = huber; |
| 37 | } |
| 38 | |
| 39 | { |
| 40 | auto gaussian = noiseModel::Isotropic::Sigma(dim: 3, sigma: 0.01); |
| 41 | auto huber = noiseModel::Robust::Create( |
| 42 | robust: noiseModel::mEstimator::Huber::Create(k: 1.345), noise: gaussian); |
| 43 | |
| 44 | velocityNoiseModel = huber; |
| 45 | } |
| 46 | |
| 47 | // expect IMU to be rotated in image space co-ords |
| 48 | auto p = std::make_shared<PreintegratedCombinedMeasurements::Params>( |
| 49 | args: Vector3(0.0, 9.8, 0.0)); |
| 50 | |
| 51 | p->accelerometerCovariance = |
| 52 | I_3x3 * pow(x: 0.0565, y: 2.0); // acc white noise in continuous |
| 53 | p->integrationCovariance = |
| 54 | I_3x3 * 1e-9; // integration uncertainty continuous |
| 55 | p->gyroscopeCovariance = |
| 56 | I_3x3 * pow(x: 4.0e-5, y: 2.0); // gyro white noise in continuous |
| 57 | p->biasAccCovariance = I_3x3 * pow(x: 0.00002, y: 2.0); // acc bias in continuous |
| 58 | p->biasOmegaCovariance = |
| 59 | I_3x3 * pow(x: 0.001, y: 2.0); // gyro bias in continuous |
| 60 | |
| 61 | #ifdef GTSAM_ALLOW_DEPRECATED_SINCE_V43 |
| 62 | p->biasAccOmegaInt = Matrix::Identity(rows: 6, cols: 6) * 1e-5; |
| 63 | #endif |
| 64 | |
| 65 | // body to IMU rotation |
| 66 | Rot3 iRb(0.036129, -0.998727, 0.035207, 0.045417, -0.033553, -0.998404, |
| 67 | 0.998315, 0.037670, 0.044147); |
| 68 | |
| 69 | // body to IMU translation (meters) |
| 70 | Point3 iTb(0.03, -0.025, -0.06); |
| 71 | |
| 72 | // body in this example is the left camera |
| 73 | p->body_P_sensor = Pose3(iRb, iTb); |
| 74 | |
| 75 | Rot3 prior_rotation = Rot3(I_3x3); |
| 76 | Pose3 prior_pose(prior_rotation, Point3(0, 0, 0)); |
| 77 | |
| 78 | Vector3 acc_bias(0.0, -0.0942015, 0.0); // in camera frame |
| 79 | Vector3 gyro_bias(-0.00527483, -0.00757152, -0.00469968); |
| 80 | |
| 81 | priorImuBias = imuBias::ConstantBias(acc_bias, gyro_bias); |
| 82 | |
| 83 | prevState = NavState(prior_pose, Vector3(0, 0, 0)); |
| 84 | propState = prevState; |
| 85 | prevBias = priorImuBias; |
| 86 | |
| 87 | preintegrated = new PreintegratedCombinedMeasurements(p, priorImuBias); |
| 88 | } |
| 89 | |
| 90 | imuBias::ConstantBias priorImuBias; // assume zero initial bias |
| 91 | noiseModel::Robust::shared_ptr velocityNoiseModel; |
| 92 | noiseModel::Robust::shared_ptr biasNoiseModel; |
| 93 | NavState prevState; |
| 94 | NavState propState; |
| 95 | imuBias::ConstantBias prevBias; |
| 96 | PreintegratedCombinedMeasurements* preintegrated; |
| 97 | }; |
| 98 | |
| 99 | int main(int argc, char* argv[]) { |
| 100 | if (argc != 2) { |
| 101 | cout << "./ISAM2_SmartFactorStereo_IMU [data.txt]\n" ; |
| 102 | return 0; |
| 103 | } |
| 104 | |
| 105 | ifstream in(argv[1]); |
| 106 | |
| 107 | if (!in) { |
| 108 | cerr << "error opening: " << argv[1] << "\n" ; |
| 109 | return 1; |
| 110 | } |
| 111 | |
| 112 | // Camera parameters |
| 113 | double fx = 822.37; |
| 114 | double fy = 822.37; |
| 115 | double cx = 538.73; |
| 116 | double cy = 579.10; |
| 117 | double baseline = 0.372; // meters |
| 118 | |
| 119 | Cal3_S2Stereo::shared_ptr K(new Cal3_S2Stereo(fx, fy, 0.0, cx, cy, baseline)); |
| 120 | |
| 121 | ISAM2Params parameters; |
| 122 | parameters.relinearizeThreshold = 0.1; |
| 123 | ISAM2 isam(parameters); |
| 124 | |
| 125 | // Create a factor graph |
| 126 | std::map<size_t, SmartStereoProjectionPoseFactor::shared_ptr> smartFactors; |
| 127 | NonlinearFactorGraph graph; |
| 128 | Values initialEstimate; |
| 129 | IMUHelper imu; |
| 130 | |
| 131 | // Pose prior - at identity |
| 132 | auto priorPoseNoise = noiseModel::Diagonal::Sigmas( |
| 133 | sigmas: (Vector(6) << Vector3::Constant(value: 0.1), Vector3::Constant(value: 0.1)).finished()); |
| 134 | graph.addPrior(key: X(j: 1), prior: Pose3::Identity(), model: priorPoseNoise); |
| 135 | initialEstimate.insert(j: X(j: 0), val: Pose3::Identity()); |
| 136 | |
| 137 | // Bias prior |
| 138 | graph.addPrior(key: B(j: 1), prior: imu.priorImuBias, model: imu.biasNoiseModel); |
| 139 | initialEstimate.insert(j: B(j: 0), val: imu.priorImuBias); |
| 140 | |
| 141 | // Velocity prior - assume stationary |
| 142 | graph.addPrior(key: V(j: 1), prior: Vector3(0, 0, 0), model: imu.velocityNoiseModel); |
| 143 | initialEstimate.insert(j: V(j: 0), val: Vector3(0, 0, 0)); |
| 144 | |
| 145 | int lastFrame = 1; |
| 146 | int frame; |
| 147 | |
| 148 | while (true) { |
| 149 | char line[1024]; |
| 150 | |
| 151 | in.getline(s: line, n: sizeof(line)); |
| 152 | stringstream ss(line); |
| 153 | char type; |
| 154 | |
| 155 | ss >> type; |
| 156 | ss >> frame; |
| 157 | |
| 158 | if (frame != lastFrame || in.eof()) { |
| 159 | cout << "Running iSAM for frame: " << lastFrame << "\n" ; |
| 160 | |
| 161 | initialEstimate.insert(j: X(j: lastFrame), val: Pose3::Identity()); |
| 162 | initialEstimate.insert(j: V(j: lastFrame), val: Vector3(0, 0, 0)); |
| 163 | initialEstimate.insert(j: B(j: lastFrame), val: imu.prevBias); |
| 164 | |
| 165 | CombinedImuFactor imuFactor(X(j: lastFrame - 1), V(j: lastFrame - 1), |
| 166 | X(j: lastFrame), V(j: lastFrame), B(j: lastFrame - 1), |
| 167 | B(j: lastFrame), *imu.preintegrated); |
| 168 | |
| 169 | graph.add(factorOrContainer: imuFactor); |
| 170 | |
| 171 | isam.update(newFactors: graph, newTheta: initialEstimate); |
| 172 | |
| 173 | Values currentEstimate = isam.calculateEstimate(); |
| 174 | |
| 175 | imu.propState = imu.preintegrated->predict(state_i: imu.prevState, bias_i: imu.prevBias); |
| 176 | imu.prevState = NavState(currentEstimate.at<Pose3>(j: X(j: lastFrame)), |
| 177 | currentEstimate.at<Vector3>(j: V(j: lastFrame))); |
| 178 | imu.prevBias = currentEstimate.at<imuBias::ConstantBias>(j: B(j: lastFrame)); |
| 179 | imu.preintegrated->resetIntegrationAndSetBias(biasHat: imu.prevBias); |
| 180 | |
| 181 | graph.resize(size: 0); |
| 182 | initialEstimate.clear(); |
| 183 | |
| 184 | if (in.eof()) { |
| 185 | break; |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | if (type == 'i') { // Process IMU measurement |
| 190 | double ax, ay, az; |
| 191 | double gx, gy, gz; |
| 192 | double dt = 1 / 800.0; // IMU at ~800Hz |
| 193 | |
| 194 | ss >> ax; |
| 195 | ss >> ay; |
| 196 | ss >> az; |
| 197 | |
| 198 | ss >> gx; |
| 199 | ss >> gy; |
| 200 | ss >> gz; |
| 201 | |
| 202 | Vector3 acc(ax, ay, az); |
| 203 | Vector3 gyr(gx, gy, gz); |
| 204 | |
| 205 | imu.preintegrated->integrateMeasurement(measuredAcc: acc, measuredOmega: gyr, dt); |
| 206 | } else if (type == 's') { // Process stereo measurement |
| 207 | int landmark; |
| 208 | double xl, xr, y; |
| 209 | |
| 210 | ss >> landmark; |
| 211 | ss >> xl; |
| 212 | ss >> xr; |
| 213 | ss >> y; |
| 214 | |
| 215 | if (smartFactors.count(x: landmark) == 0) { |
| 216 | auto gaussian = noiseModel::Isotropic::Sigma(dim: 3, sigma: 1.0); |
| 217 | |
| 218 | SmartProjectionParams params(HESSIAN, ZERO_ON_DEGENERACY); |
| 219 | |
| 220 | smartFactors[landmark] = SmartStereoProjectionPoseFactor::shared_ptr( |
| 221 | new SmartStereoProjectionPoseFactor(gaussian, params)); |
| 222 | graph.push_back(factor: smartFactors[landmark]); |
| 223 | } |
| 224 | |
| 225 | smartFactors[landmark]->add(measured: StereoPoint2(xl, xr, y), poseKey: X(j: frame), K); |
| 226 | } else { |
| 227 | throw runtime_error("unexpected data type: " + string(1, type)); |
| 228 | } |
| 229 | |
| 230 | lastFrame = frame; |
| 231 | } |
| 232 | |
| 233 | return 0; |
| 234 | } |
| 235 | |