Proposal: Add msgs to support points with covariances

Description

Currently, the geometry_msgs does only feature the Point and Point32 message to represent a point's position in 3D space. The suggestion would be to add additional messages to include the uncertainty of this position given by the 3D covariance.

Motivation

Many applications require the covariance/uncertainty of geometric information for further processing. E.g. if a detector extracts features/points of some sensor data, it may also give an uncertainty estimation on how good the detection is. This may be used in further processing.

While geometry_msgs already provides PoseWithCovariance for 6-DoF estimates, there is no canonical 3-DoF counterpart. ROS users can work around this by misusing PoseWithCovariance with a dummy orientation and a 6×6 covariance whose rotational block is meaningless — which is wasteful and semantically incorrect. The proposed messages mirror the existing pattern (Pose → PoseWithCovariance → PoseWithCovarianceStamped), making the addition a natural extension.

Design / Implementation Considerations

Add missing msgs:

• PointWithCovariance.msg
• Point32WithCovariance.msg
• PointWithCovarianceStamped.msg
• Point32WithCovarianceStamped.msg
• (Point32Stamped.msg - missing in existing representations, adding for completeness)

Use Cases

Lane boundary detection (own use case).
My research group is developing an algorithm that detects lane boundaries by extracting polylines from camera images and LiDAR scans. Merging multiple detections across sensors and time requires a 3×3 covariance per polyline vertex.

This is standard in modern lane detectors: LaneNet++: Uncertainty-Aware Lane Detection for Autonomous Vehicle (Springer Advances in Visual Computing) provides explicit uncertainty estimation on lane points to reduce false positives. Even classification-based detectors like Ultra-Fast-Lane-Detection-V2 (Qin et al., IEEE TPAMI 2022) produce a per-point softmax distribution whose expectation is already used as the point estimate — the variance is a near-zero-cost byproduct, but no standard ROS 2 message exists to transport it.

Additional Information

No response

Start by reviewing the existing geometry_msgs Point, Point32, PoseWithCovariance, and stamped message definitions to compare their naming and covariance representations. Evaluate the proposed PointWithCovariance, Point32WithCovariance, stamped variants, and Point32Stamped against those patterns and the stated 3×3 uncertainty use case. Done means an agreed message design and the complete set of interface definitions.