Merge Point Clouds

How it fits

Typical backends

Caveats

• Hard constraintInputs must already be in a common coordinate frame; there is no transform or registration here, so run alignment upstream when sensors differ. Each input's points tensor must be shaped as N rows of 3 coordinates or the call aborts.
• LimitationOverlapping regions are doubled because points are concatenated without dedup; add a decimation pass downstream if density matters.
• I/O contractColours and normals are only recognised on an input when their row count matches that input's points; otherwise they are treated as absent for that cloud.
• Parameter interactionWhen some inputs carry an optional field and others do not, drop_optional_on_mismatch either drops that field from the output entirely or zero-fills the missing rows so every cloud contributes; zero-fill yields black colours and degenerate normals that break shading.
• CostPeak memory rises to roughly the sum of all input cloud sizes during concatenation.
• State lifetimedrop_optional_on_mismatch is hot-swappable and takes effect on the next call.