# cartographer/sensor/landmark_data.h

## Overview

`LandmarkData` carries observations of fixed landmarks (visual fiducials, AprilTags, or similar) observed in the tracking frame. Each observation relates a named landmark to the robot pose with confidence weights, enabling the global optimization to anchor the map to real-world landmarks.

## Structs

### LandmarkObservation

```cpp
struct LandmarkObservation {
  std::string id;
  transform::Rigid3d landmark_to_tracking_transform;
  double translation_weight;
  double rotation_weight;
};
```

- `id` — Name of the observed landmark.
- `landmark_to_tracking_transform` — Estimated pose of the landmark in the tracking frame at observation time.
- `translation_weight` — Confidence of the translational part of the observation.
- `rotation_weight` — Confidence of the rotational part of the observation.

### LandmarkData

```cpp
struct LandmarkData {
  common::Time time;
  std::vector<LandmarkObservation> landmark_observations;
};
```

## Serialization

```cpp
proto::LandmarkData ToProto(const LandmarkData& landmark_data);
LandmarkData FromProto(const proto::LandmarkData& proto);
```

`proto::LandmarkData` (defined in `sensor/proto/sensor.proto`) stores the timestamp and repeated observations:

```protobuf
message LandmarkData {
  int64 timestamp = 1;
  repeated LandmarkObservation landmark_observations = 2;
}
```

## Usage

```cpp
sensor::LandmarkObservation observation{
    "landmark_1", landmark_to_tracking, translation_weight, rotation_weight};
sensor::LandmarkData landmark_data{time, {observation}};
trajectory_builder->AddLandmarkData(sensor_id, landmark_data);
```

## Role in the SLAM Pipeline

- **Trajectory builder interface** — `AddLandmarkData` accepts landmark observations.
- **Optimization** — The pose graph accumulates `LandmarkNode`s (a landmark's global pose estimate and per-trajectory observations). During `OptimizationProblemInterface::Solve`, landmark data is used to jointly optimize the global poses of landmarks and nodes, with the weights controlling how strongly each observation pulls the solution.
- **Serialization** — landmark data is stored per-trajectory in the serialization format via `proto::LandmarkData` in `serialization.proto`; optimized landmark poses are stored in the pose graph (`proto::PoseGraph::LandmarkPose`).

## See Also

- [Data](data.md) — Base class and `Dispatchable` wrapping
- [FixedFramePoseData](fixed_frame_pose_data.md) — Alternative absolute reference (GPS-style)
- [OptimizationProblemInterface](../mapping/internal/optimization/optimization_problem_interface.md) — Consumes landmark observations
- [PoseGraphInterface](../mapping/pose_graph_interface.md) — Defines `LandmarkNode`