Projection mapping means aligning an image to the geometry of a real object so that the two read as one thing. The technique is entirely dependent on knowing where the object is, which makes it awkward in a building whose whole business is moving objects around.
Alignment is not a setting, it is a state
A mapped surface is defined by a warp and a mask created against the object in one particular position. Move the object and both are wrong. It is not degraded gracefully: light lands past one edge onto whatever is behind, and a strip of the object that was lit goes dark.
The eye is unusually good at seeing this. A misregistration far too small to matter in most contexts reads immediately as an error, because the audience has an unambiguous reference in the physical edge sitting right next to the light.
Three ways to handle a set that moves
The first is to stop it moving during the mapped sequence. This is unglamorous, entirely reliable, and more often the right answer than production meetings tend to allow.
The second is to store an alignment for each position the unit stops in and change between them on cue. This works when the scenery arrives in the same place every time, which means driven automation with positional feedback rather than a unit pushed on by hand. Repeatability is the requirement, and it is a property of the scenic engineering rather than of the media system.
The third is to track the object continuously and warp in real time. Open protocols exist for carrying stage position data from an automation or tracking system to a media server, which is what makes this possible without proprietary coupling between departments.
Latency turns into distance
Real time tracking never arrives instantly. Position is measured, transmitted, applied to a warp, rendered and displayed, and each stage adds delay. Whatever that total is, the content is showing where the object was rather than where it is, and the error on stage is the delay multiplied by the speed of the move.
The arithmetic is unforgiving in exactly the moments a designer wants most. A slow drift hides a great deal of latency. A fast track across the stage converts a small delay into a visible offset, and it does it during the most conspicuous cue in the sequence.
Who owns the number
Mapping sits between departments that do not usually share a tolerance. Automation reports where it thinks a unit is. Scenery decides how much the unit flexes, rolls or settles once it gets there. Video decides how much error is visible. None of the three owns the whole figure, and the failure is almost always that nobody wrote it down.
A production that names the acceptable error in millimetres at the start, and asks whether the scenic engineering can meet it, has turned an argument at the technical rehearsal into a specification. A production that does not will discover the number by looking at the stage, which is the expensive way to establish it.
Content that survives being slightly wrong
The most effective mitigation is not technical. Content with soft edges, textures, gradients and no hard registration to the physical boundary tolerates a couple of centimetres of error invisibly. Content built from crisp graphic shapes that line up with the object's edges advertises every millimetre.
Choosing the forgiving version is a design decision available at the start and unavailable at the technical rehearsal, and it is worth taking before anyone has been asked whether the automation is repeatable.
The rehearsal you cannot have
Mapping is one of the few departments whose work cannot be prepared away from the object. Sound can be programmed against a recording, light can be plotted on paper, but a warp exists only in relation to a physical unit standing in a real position under a real lens. Until the scenery is built and hung, the alignment does not exist to be checked.
The scheduling consequence is that mapping consumes stage time at precisely the moment stage time is scarcest, competing with focus, with the sound check and with the cast. Productions that succeed at it generally buy the time in advance, in the form of a scenic delivery early enough to align against, rather than hoping to find it during the technical rehearsal.
The surface fights back too
Scenic paint is chosen to look right under light, not to receive an image. Gloss produces hot spots, texture breaks the image, saturated colour subtracts wavelengths, and dark surfaces absorb most of what arrives. A unit intended to be mapped needs to be specified as a projection surface at the point it is designed, which is a conversation between two departments that frequently happens too late.
Performers add the other half of the problem. Anyone between the lens and the object is a shadow on the object, and the closer they work to it the larger that shadow is. Blocking and projection have to be planned against each other rather than in sequence.
What we cannot verify
Latency figures for tracking and media systems are published by the companies selling them, are measured end to end in configurations that are not described, and change with content resolution and render load. We reproduce none of them. Positional accuracy claims for automation come from the automation supplier. The honest test is the one performed with the actual scenery, at the actual cue speed, watched from the front row where the error is largest.
The short version
- A mapped alignment is valid for one position of the object and no other.
- The physical edge gives the audience a reference that makes small errors obvious.
- Stopping the move during the sequence is reliable and underrated.
- Cued alignments need automation with positional feedback, not hand-pushed scenery.
- Tracking latency multiplied by move speed is the offset the audience sees.
- Soft-edged content forgives error; hard graphic registration advertises it.