blob: cae1c69263e64ab45377c4c238abc5e6333a5838 [file] [log] [blame]
Milind Upadhyay7c205222022-11-16 18:20:58 -08001#ifndef FRC971_VISION_TARGET_MAPPER_H_
2#define FRC971_VISION_TARGET_MAPPER_H_
3
4#include <unordered_map>
5
Austin Schuh99f7c6a2024-06-25 22:07:44 -07006#include "absl/strings/str_format.h"
Milind Upadhyayc5beba12022-12-17 17:41:20 -08007#include "ceres/ceres.h"
Philipp Schrader790cb542023-07-05 21:06:52 -07008
9#include "aos/events/simulated_event_loop.h"
Milind Upadhyay7c205222022-11-16 18:20:58 -080010#include "frc971/vision/ceres/types.h"
Milind Upadhyaycd677a32022-12-04 13:06:43 -080011#include "frc971/vision/target_map_generated.h"
milind-u8f4e43e2023-04-09 17:11:19 -070012#include "frc971/vision/visualize_robot.h"
Milind Upadhyay7c205222022-11-16 18:20:58 -080013
14namespace frc971::vision {
15
16// Estimates positions of vision targets (ex. April Tags) using
17// target detections relative to a robot (which were computed using robot
18// positions at the time of those detections). Solves SLAM problem to estimate
19// target locations using deltas between consecutive target detections.
20class TargetMapper {
21 public:
22 using TargetId = int;
Milind Upadhyayc5beba12022-12-17 17:41:20 -080023 using ConfidenceMatrix = Eigen::Matrix<double, 6, 6>;
Milind Upadhyay7c205222022-11-16 18:20:58 -080024
25 struct TargetPose {
26 TargetId id;
Milind Upadhyayc5beba12022-12-17 17:41:20 -080027 ceres::examples::Pose3d pose;
Milind Upadhyay7c205222022-11-16 18:20:58 -080028 };
29
Milind Upadhyaycd677a32022-12-04 13:06:43 -080030 // target_poses_path is the path to a TargetMap json with initial guesses for
31 // the actual locations of the targets on the field.
Milind Upadhyay7c205222022-11-16 18:20:58 -080032 // target_constraints are the deltas between consecutive target detections,
33 // and are usually prepared by the DataAdapter class below.
Milind Upadhyaycd677a32022-12-04 13:06:43 -080034 TargetMapper(std::string_view target_poses_path,
Milind Upadhyayc5beba12022-12-17 17:41:20 -080035 const ceres::examples::VectorOfConstraints &target_constraints);
Milind Upadhyaycd677a32022-12-04 13:06:43 -080036 // Alternate constructor for tests.
37 // Takes in the actual intial guesses instead of a file containing them
Milind Upadhyayc5beba12022-12-17 17:41:20 -080038 TargetMapper(const ceres::examples::MapOfPoses &target_poses,
39 const ceres::examples::VectorOfConstraints &target_constraints);
Milind Upadhyay7c205222022-11-16 18:20:58 -080040
Milind Upadhyay05652cb2022-12-07 20:51:51 -080041 // Solves for the target map. If output_dir is set, the map will be saved to
42 // output_dir/field_name.json
43 void Solve(std::string_view field_name,
44 std::optional<std::string_view> output_dir = std::nullopt);
Milind Upadhyaycd677a32022-12-04 13:06:43 -080045
46 // Prints target poses into a TargetMap flatbuffer json
Milind Upadhyay05652cb2022-12-07 20:51:51 -080047 std::string MapToJson(std::string_view field_name) const;
Milind Upadhyay7c205222022-11-16 18:20:58 -080048
49 static std::optional<TargetPose> GetTargetPoseById(
50 std::vector<TargetPose> target_poses, TargetId target_id);
51
Jim Ostrowski49be8232023-03-23 01:00:14 -070052 // Version that gets based on internal target_poses
milind-u2ab4db12023-03-25 21:59:23 -070053 std::optional<TargetPose> GetTargetPoseById(TargetId target_id) const;
Jim Ostrowski49be8232023-03-23 01:00:14 -070054
Milind Upadhyayc5beba12022-12-17 17:41:20 -080055 ceres::examples::MapOfPoses target_poses() { return target_poses_; }
Milind Upadhyay7c205222022-11-16 18:20:58 -080056
milind-u8f4e43e2023-04-09 17:11:19 -070057 // Cost function for the secondary solver finding out where the whole map fits
58 // in the world
59 template <typename S>
60 bool operator()(const S *const translation, const S *const rotation,
61 S *residual) const;
62
milind-u401de982023-04-14 17:32:03 -070063 void DumpConstraints(std::string_view path) const;
64 void DumpStats(std::string_view path) const;
Jim Ostrowskif41b0942024-03-24 18:05:02 -070065 void PrintDiffs() const;
milind-u401de982023-04-14 17:32:03 -070066
Milind Upadhyay7c205222022-11-16 18:20:58 -080067 private:
milind-u401de982023-04-14 17:32:03 -070068 // Error in an estimated pose
69 struct PoseError {
70 double angle;
71 double distance;
72 };
73
74 // Stores info on how much all the constraints differ from our solved target
75 // map
76 struct Stats {
77 // Average error for translation and rotation
78 PoseError avg_err;
79 // Standard deviation for translation and rotation error
80 PoseError std_dev;
81 // Maximum error for translation and rotation
82 PoseError max_err;
83 };
84
85 // Compute the error of a single constraint
86 PoseError ComputeError(const ceres::examples::Constraint3d &constraint) const;
87 // Compute cumulative stats for all constraints
88 Stats ComputeStats() const;
89 // Removes constraints with very large errors
90 void RemoveOutlierConstraints();
91
milind-u526d5672023-04-17 20:09:10 -070092 void CountConstraints();
93
Milind Upadhyay7c205222022-11-16 18:20:58 -080094 // Constructs the nonlinear least squares optimization problem from the
95 // pose graph constraints.
milind-u8f4e43e2023-04-09 17:11:19 -070096 void BuildTargetPoseOptimizationProblem(
Milind Upadhyayc5beba12022-12-17 17:41:20 -080097 const ceres::examples::VectorOfConstraints &constraints,
98 ceres::examples::MapOfPoses *poses, ceres::Problem *problem);
Milind Upadhyay7c205222022-11-16 18:20:58 -080099
milind-u8f4e43e2023-04-09 17:11:19 -0700100 // Constructs the nonlinear least squares optimization problem for the solved
101 // -> actual pose solver.
milind-u401de982023-04-14 17:32:03 -0700102 std::unique_ptr<ceres::CostFunction> BuildMapFittingOptimizationProblem(
103 ceres::Problem *problem);
milind-u8f4e43e2023-04-09 17:11:19 -0700104
Jim Ostrowski4527dd72024-03-07 00:20:15 -0800105 // Create and display a visualization of the graph connectivity of the
106 // constraints
107 void DisplayConstraintGraph();
108
Jim Ostrowskiefc3bde2024-03-23 16:34:06 -0700109 // Create and display a visualization of the map solution (vs. the input map)
110 void DisplaySolvedVsInitial();
111
Milind Upadhyay7c205222022-11-16 18:20:58 -0800112 // Returns true if the solve was successful.
113 bool SolveOptimizationProblem(ceres::Problem *problem);
114
milind-u8f4e43e2023-04-09 17:11:19 -0700115 ceres::examples::MapOfPoses ideal_target_poses_;
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800116 ceres::examples::MapOfPoses target_poses_;
117 ceres::examples::VectorOfConstraints target_constraints_;
milind-u8f4e43e2023-04-09 17:11:19 -0700118
milind-u526d5672023-04-17 20:09:10 -0700119 // Counts of each pair of target ids we observe, so we can scale cost based on
120 // the inverse of this and remove bias towards certain pairs
121 std::map<std::pair<TargetId, TargetId>, size_t> constraint_counts_;
122
milind-u8f4e43e2023-04-09 17:11:19 -0700123 // Transformation moving the target map we solved for to where it actually
124 // should be in the world
125 Eigen::Translation3d T_frozen_actual_;
126 Eigen::Quaterniond R_frozen_actual_;
127
Jim Ostrowski68e56172023-09-17 23:44:15 -0700128 const double kFieldWidth_ = 20.0; // 20 meters across
129 const int kImageWidth_ = 1000;
130 const int kImageHeight_ =
131 kImageWidth_ * 3.0 / 4.0; // Roughly matches field aspect ratio
milind-u8f4e43e2023-04-09 17:11:19 -0700132 mutable VisualizeRobot vis_robot_;
milind-u401de982023-04-14 17:32:03 -0700133
134 Stats stats_with_outliers_;
Milind Upadhyay7c205222022-11-16 18:20:58 -0800135};
136
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800137// Utility functions for dealing with ceres::examples::Pose3d structs
Milind Upadhyay7c205222022-11-16 18:20:58 -0800138class PoseUtils {
139 public:
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800140 // Embeds a 3d pose into an affine transformation
141 static Eigen::Affine3d Pose3dToAffine3d(
142 const ceres::examples::Pose3d &pose3d);
143 // Inverse of above function
144 static ceres::examples::Pose3d Affine3dToPose3d(const Eigen::Affine3d &H);
Milind Upadhyay7c205222022-11-16 18:20:58 -0800145
146 // Computes pose_2 relative to pose_1. This is equivalent to (pose_1^-1 *
147 // pose_2)
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800148 static ceres::examples::Pose3d ComputeRelativePose(
149 const ceres::examples::Pose3d &pose_1,
150 const ceres::examples::Pose3d &pose_2);
Milind Upadhyay7c205222022-11-16 18:20:58 -0800151
152 // Computes pose_2 given a pose_1 and pose_2 relative to pose_1. This is
153 // equivalent to (pose_1 * pose_2_relative)
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800154 static ceres::examples::Pose3d ComputeOffsetPose(
155 const ceres::examples::Pose3d &pose_1,
156 const ceres::examples::Pose3d &pose_2_relative);
157
158 // Converts a rotation with roll, pitch, and yaw into a quaternion
159 static Eigen::Quaterniond EulerAnglesToQuaternion(const Eigen::Vector3d &rpy);
160 // Inverse of above function
161 static Eigen::Vector3d QuaternionToEulerAngles(const Eigen::Quaterniond &q);
162 // Converts a 3d rotation matrix into a rotation with roll, pitch, and yaw
163 static Eigen::Vector3d RotationMatrixToEulerAngles(const Eigen::Matrix3d &R);
milind-u3f5f83c2023-01-29 15:23:51 -0800164
165 // Builds a TargetPoseFbs from a TargetPose
166 static flatbuffers::Offset<TargetPoseFbs> TargetPoseToFbs(
167 const TargetMapper::TargetPose &target_pose,
168 flatbuffers::FlatBufferBuilder *fbb);
169 // Converts a TargetPoseFbs to a TargetPose
170 static TargetMapper::TargetPose TargetPoseFromFbs(
171 const TargetPoseFbs &target_pose_fbs);
Milind Upadhyay7c205222022-11-16 18:20:58 -0800172};
173
174// Transforms robot position and target detection data into target constraints
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800175// to be used for mapping.
Milind Upadhyay7c205222022-11-16 18:20:58 -0800176class DataAdapter {
177 public:
Milind Upadhyay7c205222022-11-16 18:20:58 -0800178 // Pairs target detection with a time point
179 struct TimestampedDetection {
180 aos::distributed_clock::time_point time;
181 // Pose of target relative to robot
182 Eigen::Affine3d H_robot_target;
Milind Upadhyayebf93ee2023-01-05 14:12:58 -0800183 // Horizontal distance from camera to target, used for confidence
184 // calculation
185 double distance_from_camera;
milind-ud62f80a2023-03-04 16:37:09 -0800186 // A measure of how much distortion affected this detection from 0-1.
187 double distortion_factor;
Milind Upadhyay7c205222022-11-16 18:20:58 -0800188 TargetMapper::TargetId id;
189 };
190
Milind Upadhyayec493912022-12-18 21:33:15 -0800191 // Pairs consecutive target detections that are not too far apart in time into
192 // constraints. Meant to be used on a system without a position measurement.
193 // Assumes timestamped_target_detections is in chronological order.
194 // max_dt is the maximum time between two target detections to match them up.
195 // If too much time passes, the recoding device (box of pis) could have moved
196 // too much
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800197 static ceres::examples::VectorOfConstraints MatchTargetDetections(
Milind Upadhyayec493912022-12-18 21:33:15 -0800198 const std::vector<TimestampedDetection> &timestamped_target_detections,
milind-u8791bed2023-03-04 14:45:52 -0800199 aos::distributed_clock::duration max_dt = std::chrono::milliseconds(10));
Milind Upadhyayec493912022-12-18 21:33:15 -0800200
Milind Upadhyay7c205222022-11-16 18:20:58 -0800201 // Computes inverse of covariance matrix, assuming there was a target
202 // detection between robot movement over the given time period. Ceres calls
203 // this matrix the "information"
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800204 static TargetMapper::ConfidenceMatrix ComputeConfidence(
milind-ud62f80a2023-03-04 16:37:09 -0800205 const TimestampedDetection &detection_start,
206 const TimestampedDetection &detection_end);
Milind Upadhyay7c205222022-11-16 18:20:58 -0800207
Milind Upadhyay7c205222022-11-16 18:20:58 -0800208 // Computes the constraint between the start and end pose of the targets: the
209 // relative pose between the start and end target locations in the frame of
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800210 // the start target.
211 static ceres::examples::Constraint3d ComputeTargetConstraint(
Milind Upadhyayec493912022-12-18 21:33:15 -0800212 const TimestampedDetection &target_detection_start,
213 const TimestampedDetection &target_detection_end,
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800214 const TargetMapper::ConfidenceMatrix &confidence);
Milind Upadhyay7c205222022-11-16 18:20:58 -0800215};
216
milind-ufbc5c812023-04-06 21:24:29 -0700217} // namespace frc971::vision
218
Austin Schuh99f7c6a2024-06-25 22:07:44 -0700219namespace ceres::examples {
220template <typename Sink>
221void AbslStringify(Sink &sink, ceres::examples::Pose3d pose) {
222 auto rpy = frc971::vision::PoseUtils::QuaternionToEulerAngles(pose.q);
223 absl::Format(&sink,
224 "{x: %.3f, y: %.3f, z: %.3f, roll: %.3f, pitch: "
225 "%.3f, yaw: %.3f}",
226 pose.p(0), pose.p(1), pose.p(2), rpy(0), rpy(1), rpy(2));
227}
228
229template <typename Sink>
230void AbslStringify(Sink &sink, ceres::examples::Constraint3d constraint) {
231 absl::Format(&sink, "{id_begin: %d, id_end: %d, pose: ", constraint.id_begin,
232 constraint.id_end);
233 AbslStringify(sink, constraint.t_be);
234 sink.Append("}");
235}
236} // namespace ceres::examples
Milind Upadhyayc5beba12022-12-17 17:41:20 -0800237
Milind Upadhyay7c205222022-11-16 18:20:58 -0800238#endif // FRC971_VISION_TARGET_MAPPER_H_