Releasing the Absent Ghosts

2026 / INTERACTIVE INSTALLATION

EXHIBITIONV&A Digital Design Weekend

LOCATIONSouth Kensington, London

DATE18–20 September 2026

3D scanning · Computational assemblage · Real-time environments · Physical computing

Project lead · Research & installation design · Algorithm development · Unity / HDRP · Environment-transition shader

Releasing the Absent Ghosts examines how temporary structures and traces of everyday life are overlooked as cities change. Scanned and sourced objects are assembled inside an AI-generated architectural form.

Visitors turn a weighted hand crank to open the digital monument, reveal the objects within it and transform the surrounding landscape. The installation asks how interaction can bring overlooked objects and structures back into view.

A familiar playground gives way to a shifting landscape. Continued turning opens the monument, bringing its individual objects into view.

Research & development

Releasing the Absent Ghosts / From urban traces to a hand-cranked monument

View processShow less

Fieldwork in Yongqingfang and Zhujiang Paper Mill, Guangzhou, focused on temporary structures and everyday traces often overlooked as cities change. The installation draws on these observations, combining scanned objects and sourced models within an AI-generated architectural form.

01 / RECORD

Recording objects in 3D

Scanning captures an object's shape and surface so it can be used in the digital assembly. The models retain visible details of use, but do not reproduce the object's original setting.

Output: 3D models for assembly.

02 / DEFINE

Setting the architectural form

An AI-generated architectural model defines the volume to be filled. Its monumental scale contrasts with the temporary structures and everyday objects gathered for the project. It is a design proposal, not a reconstruction of either site.

Output: a target volume for object placement.

03 / ASSEMBLE

Keeping the objects separate

The algorithm places individual models inside the target volume while retaining each object's geometry and identity. Together they form a building-like silhouette; when the exterior opens, viewers can see the objects that make it up.

Output: an assembly of individually identifiable objects.

Early placement tests explored paths and growth fields. The final approach fills an AI-generated architectural volume with individual objects, preserving their separate geometry for use in Unity.

Object placement development study — Early studies / paths → growth fields → solid occupancy

Assemblage of individual objects — Assemblage / AI-assisted reconstruction

The solver groups objects as linear, planar or volumetric, then controls placement through volume and overlap limits. Each instance retains its source ID, position, rotation and scale for export to Unity.

200 source models → 865 instances · 87.7% fill

11.8% overlap · 0.0604% outside · 58 min

Single production run; not a comparative benchmark.

Method references: Spectral Packing — Cui et al. (2023); design-led 3D packing — Gerritsen et al. (2022).

Searching for Object Placements

For each sampled rotation and scale, FFT correlation scores possible positions:

C_A(t) = Σₓ A(t + x)F(x)

F is the object's voxel mask; A is a spatial field; t is translation. Placements are filtered by overlap and boundary limits. The solver accepts the best valid candidate, updates the occupied space and repeats.

Target mesh, 865 instance centres and object poses

In Unity, a foggy playground gives the monument a familiar urban setting. Turning the crank opens its exterior, reveals the objects inside and changes the surrounding landscape.

Initial playground and monument scene — Initial scene / playground and monument → Environment transition

Transformed environment — Initial scene / playground and monument → Environment transition

The monument opens through fracture, fragment motion and dissolve. A shared progress value also drives the landscape transition, with custom Gaussian-splat rendering controls for dispersal, scale and opacity.

Fracture, opening and dispersal geometry study — Geometry studies / fracture, opening and dispersal / Unity ShaderGraph

A weighted film crank controls the transformation: keep turning to open the monument; pause to let it close after a delay. Adjustable mechanical resistance makes revealing the objects a sustained physical action.

Physical assembly / dimensions from supplied drawing · Scroll sideways to inspect the diagram

Hall sensor / rotation sensing

A disc-rim magnet triggers a Hall sensor. Arduino sends pulses to Unity over USB; pulse timing estimates rotation speed and advances the sequence. The sensor does not measure force.

Magnet → Hall sensor → Arduino → USB serial → Unity

Printed components

PLA components were 3D printed, sanded, primed, painted and weathered.

Supporting frame

Printed components are assembled around a supporting frame.

Informal Visitor Feedback

Visitors commented positively on the crank's tactile feel and the connection between turning and the projected changes. Some expressed surprise when they realised they were controlling the transformation.

Design Reflection

The weighted film crank was chosen to make continued turning require effort. This gives visitors an active role in revealing the objects. The feedback supports the choice of interface, but does not establish how visitors understood the project's account of urban change.

Next Research Question

How does the crank's resistance affect visitors' sense of control and their attention to the revealed objects? A follow-up study could compare resistance settings and ask visitors what they noticed and understood.

Visitor operating the hand-cranked installation