Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 1 | #include "frc971/vision/target_mapper.h" |
| 2 | |
| 3 | #include "frc971/control_loops/control_loop.h" |
| 4 | #include "frc971/vision/ceres/angle_local_parameterization.h" |
| 5 | #include "frc971/vision/ceres/normalize_angle.h" |
| 6 | #include "frc971/vision/ceres/pose_graph_2d_error_term.h" |
| 7 | #include "frc971/vision/geometry.h" |
| 8 | |
| 9 | DEFINE_uint64(max_num_iterations, 100, |
| 10 | "Maximum number of iterations for the ceres solver"); |
| 11 | |
| 12 | namespace frc971::vision { |
| 13 | |
| 14 | Eigen::Affine3d PoseUtils::Pose2dToAffine3d(ceres::examples::Pose2d pose2d) { |
| 15 | Eigen::Affine3d H_world_pose = |
| 16 | Eigen::Translation3d(pose2d.x, pose2d.y, 0.0) * |
| 17 | Eigen::AngleAxisd(pose2d.yaw_radians, Eigen::Vector3d::UnitZ()); |
| 18 | return H_world_pose; |
| 19 | } |
| 20 | |
| 21 | ceres::examples::Pose2d PoseUtils::Affine3dToPose2d(Eigen::Affine3d H) { |
| 22 | Eigen::Vector3d T = H.translation(); |
| 23 | double heading = std::atan2(H.rotation()(1, 0), H.rotation()(0, 0)); |
| 24 | return ceres::examples::Pose2d{T[0], T[1], |
| 25 | ceres::examples::NormalizeAngle(heading)}; |
| 26 | } |
| 27 | |
| 28 | ceres::examples::Pose2d PoseUtils::ComputeRelativePose( |
| 29 | ceres::examples::Pose2d pose_1, ceres::examples::Pose2d pose_2) { |
| 30 | Eigen::Affine3d H_world_1 = Pose2dToAffine3d(pose_1); |
| 31 | Eigen::Affine3d H_world_2 = Pose2dToAffine3d(pose_2); |
| 32 | |
| 33 | // Get the location of 2 in the 1 frame |
| 34 | Eigen::Affine3d H_1_2 = H_world_1.inverse() * H_world_2; |
| 35 | return Affine3dToPose2d(H_1_2); |
| 36 | } |
| 37 | |
| 38 | ceres::examples::Pose2d PoseUtils::ComputeOffsetPose( |
| 39 | ceres::examples::Pose2d pose_1, ceres::examples::Pose2d pose_2_relative) { |
| 40 | auto H_world_1 = Pose2dToAffine3d(pose_1); |
| 41 | auto H_1_2 = Pose2dToAffine3d(pose_2_relative); |
| 42 | auto H_world_2 = H_world_1 * H_1_2; |
| 43 | |
| 44 | return Affine3dToPose2d(H_world_2); |
| 45 | } |
| 46 | |
| 47 | namespace { |
| 48 | double ExponentiatedSinTerm(double theta) { |
| 49 | return (theta == 0.0 ? 1.0 : std::sin(theta) / theta); |
| 50 | } |
| 51 | |
| 52 | double ExponentiatedCosTerm(double theta) { |
| 53 | return (theta == 0.0 ? 0.0 : (1 - std::cos(theta)) / theta); |
| 54 | } |
| 55 | } // namespace |
| 56 | |
| 57 | ceres::examples::Pose2d DataAdapter::InterpolatePose( |
| 58 | const TimestampedPose &pose_start, const TimestampedPose &pose_end, |
| 59 | aos::distributed_clock::time_point time) { |
| 60 | auto delta_pose = |
| 61 | PoseUtils::ComputeRelativePose(pose_start.pose, pose_end.pose); |
| 62 | // Time from start of period, on the scale 0-1 where 1 is the end. |
| 63 | double interpolation_scalar = |
| 64 | static_cast<double>((time - pose_start.time).count()) / |
| 65 | static_cast<double>((pose_end.time - pose_start.time).count()); |
| 66 | |
| 67 | double theta = delta_pose.yaw_radians; |
| 68 | // Take the log of the transformation matrix: |
| 69 | // https://mathoverflow.net/questions/118533/how-to-compute-se2-group-exponential-and-logarithm |
| 70 | StdFormLine dx_line = {.a = ExponentiatedSinTerm(theta), |
| 71 | .b = -ExponentiatedCosTerm(theta), |
| 72 | .c = delta_pose.x}; |
| 73 | StdFormLine dy_line = {.a = ExponentiatedCosTerm(theta), |
| 74 | .b = ExponentiatedSinTerm(theta), |
| 75 | .c = delta_pose.y}; |
| 76 | |
| 77 | std::optional<cv::Point2d> solution = dx_line.Intersection(dy_line); |
| 78 | CHECK(solution.has_value()); |
| 79 | |
| 80 | // Re-exponentiate with the new values scaled by the interpolation scalar to |
| 81 | // get an interpolated tranformation matrix |
| 82 | double a = solution->x * interpolation_scalar; |
| 83 | double b = solution->y * interpolation_scalar; |
| 84 | double alpha = theta * interpolation_scalar; |
| 85 | |
| 86 | ceres::examples::Pose2d interpolated_pose = { |
| 87 | .x = a * ExponentiatedSinTerm(theta) - b * ExponentiatedCosTerm(theta), |
| 88 | .y = a * ExponentiatedCosTerm(theta) + b * ExponentiatedSinTerm(theta), |
| 89 | .yaw_radians = alpha}; |
| 90 | |
| 91 | return PoseUtils::ComputeOffsetPose(pose_start.pose, interpolated_pose); |
| 92 | } // namespace frc971::vision |
| 93 | |
| 94 | std::pair<std::vector<ceres::examples::Constraint2d>, |
| 95 | std::vector<ceres::examples::Pose2d>> |
| 96 | DataAdapter::MatchTargetDetections( |
| 97 | const std::vector<TimestampedPose> ×tamped_robot_poses, |
| 98 | const std::vector<TimestampedDetection> ×tamped_target_detections) { |
| 99 | // Interpolate robot poses |
| 100 | std::map<aos::distributed_clock::time_point, ceres::examples::Pose2d> |
| 101 | interpolated_poses; |
| 102 | |
| 103 | CHECK_GT(timestamped_robot_poses.size(), 1ul) |
| 104 | << "Need more than 1 robot pose"; |
| 105 | auto robot_pose_it = timestamped_robot_poses.begin(); |
| 106 | for (const auto ×tamped_detection : timestamped_target_detections) { |
| 107 | aos::distributed_clock::time_point target_time = timestamped_detection.time; |
Milind Upadhyay | 81420a2 | 2022-12-14 21:13:06 -0800 | [diff] [blame] | 108 | |
| 109 | // Skip this target detection if we have no robot poses before it |
| 110 | if (robot_pose_it->time > target_time) { |
| 111 | continue; |
| 112 | } |
| 113 | |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 114 | // Find the robot point right before this localization |
| 115 | while (robot_pose_it->time > target_time || |
| 116 | (robot_pose_it + 1)->time <= target_time) { |
| 117 | robot_pose_it++; |
| 118 | CHECK(robot_pose_it < timestamped_robot_poses.end() - 1) |
| 119 | << "Need a robot pose before and after every target detection"; |
| 120 | } |
| 121 | |
| 122 | auto start = robot_pose_it; |
| 123 | auto end = robot_pose_it + 1; |
| 124 | interpolated_poses.emplace(target_time, |
| 125 | InterpolatePose(*start, *end, target_time)); |
| 126 | } |
| 127 | |
Milind Upadhyay | 81420a2 | 2022-12-14 21:13:06 -0800 | [diff] [blame] | 128 | // In the case that all target detections were before the first robot |
| 129 | // detection, we would have no interpolated poses at this point |
| 130 | CHECK_GT(interpolated_poses.size(), 0ul) |
| 131 | << "Need a robot pose before and after every target detection"; |
| 132 | |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 133 | // Match consecutive detections |
| 134 | std::vector<ceres::examples::Constraint2d> target_constraints; |
| 135 | std::vector<ceres::examples::Pose2d> robot_delta_poses; |
| 136 | |
| 137 | auto last_detection = timestamped_target_detections[0]; |
| 138 | auto last_robot_pose = |
| 139 | interpolated_poses[timestamped_target_detections[0].time]; |
| 140 | |
| 141 | for (auto it = timestamped_target_detections.begin() + 1; |
| 142 | it < timestamped_target_detections.end(); it++) { |
| 143 | // Skip two consecutive detections of the same target, because the solver |
| 144 | // doesn't allow this |
| 145 | if (it->id == last_detection.id) { |
| 146 | continue; |
| 147 | } |
| 148 | |
| 149 | auto robot_pose = interpolated_poses[it->time]; |
| 150 | auto robot_delta_pose = |
| 151 | PoseUtils::ComputeRelativePose(last_robot_pose, robot_pose); |
| 152 | auto confidence = ComputeConfidence(last_detection.time, it->time); |
| 153 | |
| 154 | target_constraints.emplace_back(ComputeTargetConstraint( |
| 155 | last_detection, PoseUtils::Pose2dToAffine3d(robot_delta_pose), *it, |
| 156 | confidence)); |
| 157 | robot_delta_poses.emplace_back(robot_delta_pose); |
| 158 | |
| 159 | last_detection = *it; |
| 160 | last_robot_pose = robot_pose; |
| 161 | } |
| 162 | |
| 163 | return {target_constraints, robot_delta_poses}; |
| 164 | } |
| 165 | |
Milind Upadhyay | ec49391 | 2022-12-18 21:33:15 -0800 | [diff] [blame^] | 166 | std::vector<ceres::examples::Constraint2d> DataAdapter::MatchTargetDetections( |
| 167 | const std::vector<DataAdapter::TimestampedDetection> |
| 168 | ×tamped_target_detections, |
| 169 | aos::distributed_clock::duration max_dt) { |
| 170 | CHECK_GE(timestamped_target_detections.size(), 2ul) |
| 171 | << "Must have at least 2 detections"; |
| 172 | |
| 173 | // Match consecutive detections |
| 174 | std::vector<ceres::examples::Constraint2d> target_constraints; |
| 175 | for (auto it = timestamped_target_detections.begin() + 1; |
| 176 | it < timestamped_target_detections.end(); it++) { |
| 177 | auto last_detection = *(it - 1); |
| 178 | |
| 179 | // Skip two consecutive detections of the same target, because the solver |
| 180 | // doesn't allow this |
| 181 | if (it->id == last_detection.id) { |
| 182 | continue; |
| 183 | } |
| 184 | |
| 185 | // Don't take into account constraints too far apart in time, because the |
| 186 | // recording device could have moved too much |
| 187 | if ((it->time - last_detection.time) > max_dt) { |
| 188 | continue; |
| 189 | } |
| 190 | |
| 191 | // TODO(milind): better way to compute confidence since these detections are |
| 192 | // likely very close in time together |
| 193 | auto confidence = ComputeConfidence(last_detection.time, it->time); |
| 194 | target_constraints.emplace_back( |
| 195 | ComputeTargetConstraint(last_detection, *it, confidence)); |
| 196 | } |
| 197 | |
| 198 | return target_constraints; |
| 199 | } |
| 200 | |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 201 | Eigen::Matrix3d DataAdapter::ComputeConfidence( |
| 202 | aos::distributed_clock::time_point start, |
| 203 | aos::distributed_clock::time_point end) { |
| 204 | constexpr size_t kX = 0; |
| 205 | constexpr size_t kY = 1; |
| 206 | constexpr size_t kTheta = 2; |
| 207 | |
| 208 | // Uncertainty matrix between start and end |
| 209 | Eigen::Matrix3d P = Eigen::Matrix3d::Zero(); |
| 210 | |
| 211 | { |
| 212 | // Noise for odometry-based robot position measurements |
| 213 | Eigen::Matrix3d Q_odometry = Eigen::Matrix3d::Zero(); |
| 214 | Q_odometry(kX, kX) = std::pow(0.045, 2); |
| 215 | Q_odometry(kY, kY) = std::pow(0.045, 2); |
| 216 | Q_odometry(kTheta, kTheta) = std::pow(0.01, 2); |
| 217 | |
| 218 | // Add uncertainty for robot position measurements from start to end |
| 219 | int iterations = (end - start) / frc971::controls::kLoopFrequency; |
| 220 | P += static_cast<double>(iterations) * Q_odometry; |
| 221 | } |
| 222 | |
| 223 | { |
| 224 | // Noise for vision-based target localizations |
| 225 | Eigen::Matrix3d Q_vision = Eigen::Matrix3d::Zero(); |
| 226 | Q_vision(kX, kX) = std::pow(0.09, 2); |
| 227 | Q_vision(kY, kY) = std::pow(0.09, 2); |
| 228 | Q_vision(kTheta, kTheta) = std::pow(0.02, 2); |
| 229 | |
| 230 | // Add uncertainty for the 2 vision measurements (1 at start and 1 at end) |
| 231 | P += 2.0 * Q_vision; |
| 232 | } |
| 233 | |
| 234 | return P.inverse(); |
| 235 | } |
| 236 | |
| 237 | ceres::examples::Constraint2d DataAdapter::ComputeTargetConstraint( |
| 238 | const TimestampedDetection &target_detection_start, |
| 239 | const Eigen::Affine3d &H_robotstart_robotend, |
| 240 | const TimestampedDetection &target_detection_end, |
| 241 | const Eigen::Matrix3d &confidence) { |
| 242 | // Compute the relative pose (constraint) between the two targets |
| 243 | Eigen::Affine3d H_targetstart_targetend = |
| 244 | target_detection_start.H_robot_target.inverse() * H_robotstart_robotend * |
| 245 | target_detection_end.H_robot_target; |
| 246 | ceres::examples::Pose2d target_constraint = |
| 247 | PoseUtils::Affine3dToPose2d(H_targetstart_targetend); |
| 248 | |
| 249 | return ceres::examples::Constraint2d{ |
| 250 | target_detection_start.id, target_detection_end.id, |
| 251 | target_constraint.x, target_constraint.y, |
| 252 | target_constraint.yaw_radians, confidence}; |
| 253 | } |
| 254 | |
Milind Upadhyay | ec49391 | 2022-12-18 21:33:15 -0800 | [diff] [blame^] | 255 | ceres::examples::Constraint2d DataAdapter::ComputeTargetConstraint( |
| 256 | const TimestampedDetection &target_detection_start, |
| 257 | const TimestampedDetection &target_detection_end, |
| 258 | const Eigen::Matrix3d &confidence) { |
| 259 | return ComputeTargetConstraint(target_detection_start, |
| 260 | Eigen::Affine3d(Eigen::Matrix4d::Identity()), |
| 261 | target_detection_end, confidence); |
| 262 | } |
| 263 | |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 264 | TargetMapper::TargetMapper( |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 265 | std::string_view target_poses_path, |
| 266 | std::vector<ceres::examples::Constraint2d> target_constraints) |
| 267 | : target_constraints_(target_constraints) { |
| 268 | aos::FlatbufferDetachedBuffer<TargetMap> target_map = |
| 269 | aos::JsonFileToFlatbuffer<TargetMap>(target_poses_path); |
| 270 | for (const auto *target_pose_fbs : *target_map.message().target_poses()) { |
| 271 | target_poses_[target_pose_fbs->id()] = ceres::examples::Pose2d{ |
| 272 | target_pose_fbs->x(), target_pose_fbs->y(), target_pose_fbs->yaw()}; |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | TargetMapper::TargetMapper( |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 277 | std::map<TargetId, ceres::examples::Pose2d> target_poses, |
| 278 | std::vector<ceres::examples::Constraint2d> target_constraints) |
| 279 | : target_poses_(target_poses), target_constraints_(target_constraints) {} |
| 280 | |
| 281 | std::optional<TargetMapper::TargetPose> TargetMapper::GetTargetPoseById( |
| 282 | std::vector<TargetMapper::TargetPose> target_poses, TargetId target_id) { |
| 283 | for (auto target_pose : target_poses) { |
| 284 | if (target_pose.id == target_id) { |
| 285 | return target_pose; |
| 286 | } |
| 287 | } |
| 288 | |
| 289 | return std::nullopt; |
| 290 | } |
| 291 | |
| 292 | // Taken from ceres/examples/slam/pose_graph_2d/pose_graph_2d.cc |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 293 | void TargetMapper::BuildOptimizationProblem( |
| 294 | std::map<int, ceres::examples::Pose2d> *poses, |
| 295 | const std::vector<ceres::examples::Constraint2d> &constraints, |
| 296 | ceres::Problem *problem) { |
| 297 | CHECK_NOTNULL(poses); |
| 298 | CHECK_NOTNULL(problem); |
| 299 | if (constraints.empty()) { |
| 300 | LOG(WARNING) << "No constraints, no problem to optimize."; |
| 301 | return; |
| 302 | } |
| 303 | |
| 304 | ceres::LossFunction *loss_function = new ceres::HuberLoss(2.0); |
| 305 | ceres::LocalParameterization *angle_local_parameterization = |
| 306 | ceres::examples::AngleLocalParameterization::Create(); |
| 307 | |
| 308 | for (std::vector<ceres::examples::Constraint2d>::const_iterator |
| 309 | constraints_iter = constraints.begin(); |
| 310 | constraints_iter != constraints.end(); ++constraints_iter) { |
| 311 | const ceres::examples::Constraint2d &constraint = *constraints_iter; |
| 312 | |
| 313 | std::map<int, ceres::examples::Pose2d>::iterator pose_begin_iter = |
| 314 | poses->find(constraint.id_begin); |
| 315 | CHECK(pose_begin_iter != poses->end()) |
| 316 | << "Pose with ID: " << constraint.id_begin << " not found."; |
| 317 | std::map<int, ceres::examples::Pose2d>::iterator pose_end_iter = |
| 318 | poses->find(constraint.id_end); |
| 319 | CHECK(pose_end_iter != poses->end()) |
| 320 | << "Pose with ID: " << constraint.id_end << " not found."; |
| 321 | |
| 322 | const Eigen::Matrix3d sqrt_information = |
| 323 | constraint.information.llt().matrixL(); |
| 324 | // Ceres will take ownership of the pointer. |
| 325 | ceres::CostFunction *cost_function = |
| 326 | ceres::examples::PoseGraph2dErrorTerm::Create( |
| 327 | constraint.x, constraint.y, constraint.yaw_radians, |
| 328 | sqrt_information); |
| 329 | problem->AddResidualBlock( |
| 330 | cost_function, loss_function, &pose_begin_iter->second.x, |
| 331 | &pose_begin_iter->second.y, &pose_begin_iter->second.yaw_radians, |
| 332 | &pose_end_iter->second.x, &pose_end_iter->second.y, |
| 333 | &pose_end_iter->second.yaw_radians); |
| 334 | |
| 335 | problem->SetParameterization(&pose_begin_iter->second.yaw_radians, |
| 336 | angle_local_parameterization); |
| 337 | problem->SetParameterization(&pose_end_iter->second.yaw_radians, |
| 338 | angle_local_parameterization); |
| 339 | } |
| 340 | |
| 341 | // The pose graph optimization problem has three DOFs that are not fully |
| 342 | // constrained. This is typically referred to as gauge freedom. You can apply |
| 343 | // a rigid body transformation to all the nodes and the optimization problem |
| 344 | // will still have the exact same cost. The Levenberg-Marquardt algorithm has |
| 345 | // internal damping which mitigates this issue, but it is better to properly |
| 346 | // constrain the gauge freedom. This can be done by setting one of the poses |
| 347 | // as constant so the optimizer cannot change it. |
| 348 | std::map<int, ceres::examples::Pose2d>::iterator pose_start_iter = |
| 349 | poses->begin(); |
| 350 | CHECK(pose_start_iter != poses->end()) << "There are no poses."; |
| 351 | problem->SetParameterBlockConstant(&pose_start_iter->second.x); |
| 352 | problem->SetParameterBlockConstant(&pose_start_iter->second.y); |
| 353 | problem->SetParameterBlockConstant(&pose_start_iter->second.yaw_radians); |
| 354 | } |
| 355 | |
| 356 | // Taken from ceres/examples/slam/pose_graph_2d/pose_graph_2d.cc |
| 357 | bool TargetMapper::SolveOptimizationProblem(ceres::Problem *problem) { |
| 358 | CHECK_NOTNULL(problem); |
| 359 | |
| 360 | ceres::Solver::Options options; |
| 361 | options.max_num_iterations = FLAGS_max_num_iterations; |
| 362 | options.linear_solver_type = ceres::SPARSE_NORMAL_CHOLESKY; |
| 363 | |
| 364 | ceres::Solver::Summary summary; |
| 365 | ceres::Solve(options, problem, &summary); |
| 366 | |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 367 | LOG(INFO) << summary.FullReport() << '\n'; |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 368 | |
| 369 | return summary.IsSolutionUsable(); |
| 370 | } |
| 371 | |
Milind Upadhyay | 05652cb | 2022-12-07 20:51:51 -0800 | [diff] [blame] | 372 | void TargetMapper::Solve(std::string_view field_name, |
| 373 | std::optional<std::string_view> output_dir) { |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 374 | ceres::Problem problem; |
| 375 | BuildOptimizationProblem(&target_poses_, target_constraints_, &problem); |
| 376 | |
| 377 | CHECK(SolveOptimizationProblem(&problem)) |
| 378 | << "The solve was not successful, exiting."; |
| 379 | |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 380 | // TODO(milind): add origin to first target offset to all poses |
| 381 | |
Milind Upadhyay | 05652cb | 2022-12-07 20:51:51 -0800 | [diff] [blame] | 382 | auto map_json = MapToJson(field_name); |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 383 | VLOG(1) << "Solved target poses: " << map_json; |
Milind Upadhyay | 05652cb | 2022-12-07 20:51:51 -0800 | [diff] [blame] | 384 | |
| 385 | if (output_dir.has_value()) { |
| 386 | std::string output_path = |
| 387 | absl::StrCat(output_dir.value(), "/", field_name, ".json"); |
| 388 | LOG(INFO) << "Writing map to file: " << output_path; |
| 389 | aos::util::WriteStringToFileOrDie(output_path, map_json); |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 390 | } |
| 391 | } |
| 392 | |
Milind Upadhyay | 05652cb | 2022-12-07 20:51:51 -0800 | [diff] [blame] | 393 | std::string TargetMapper::MapToJson(std::string_view field_name) const { |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 394 | flatbuffers::FlatBufferBuilder fbb; |
| 395 | |
| 396 | // Convert poses to flatbuffers |
| 397 | std::vector<flatbuffers::Offset<TargetPoseFbs>> target_poses_fbs; |
| 398 | for (const auto &[id, pose] : target_poses_) { |
| 399 | TargetPoseFbs::Builder target_pose_builder(fbb); |
| 400 | target_pose_builder.add_id(id); |
Milind Upadhyay | cf75777 | 2022-12-14 20:57:36 -0800 | [diff] [blame] | 401 | target_pose_builder.add_x(pose.x); |
| 402 | target_pose_builder.add_y(pose.y); |
| 403 | target_pose_builder.add_yaw(pose.yaw_radians); |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 404 | |
| 405 | target_poses_fbs.emplace_back(target_pose_builder.Finish()); |
| 406 | } |
| 407 | |
Milind Upadhyay | 05652cb | 2022-12-07 20:51:51 -0800 | [diff] [blame] | 408 | const auto field_name_offset = fbb.CreateString(field_name); |
| 409 | flatbuffers::Offset<TargetMap> target_map_offset = CreateTargetMap( |
| 410 | fbb, fbb.CreateVector(target_poses_fbs), field_name_offset); |
Milind Upadhyay | cd677a3 | 2022-12-04 13:06:43 -0800 | [diff] [blame] | 411 | |
| 412 | return aos::FlatbufferToJson( |
| 413 | flatbuffers::GetMutableTemporaryPointer(fbb, target_map_offset), |
| 414 | {.multi_line = true}); |
Milind Upadhyay | 7c20522 | 2022-11-16 18:20:58 -0800 | [diff] [blame] | 415 | } |
| 416 | |
| 417 | } // namespace frc971::vision |