Technical Documentation

System Overview

The null² pavilion comprises five primary technology systems: the Mirror Membrane, Robotic Actuation, Digital Twin, Sensing Network, and Subwoofer Vibration System. These systems do not operate in isolation; they are interconnected in real-time and function as an integrated system that responds to visitor presence and movement.

This document provides engineer-focused specifications, design philosophy, and engineering challenges and solutions for each system. null² is not merely an installation; it is an engineering project that fuses architecture, robotics, computing, and acoustic engineering into a unified experiential space.

1. Mirror Membrane System

1.1 Material Specifications

The mirror membrane that forms the exterior and interior walls of null² is the heart of the pavilion's visual experience. It achieves 98% reflectivity through a precision mirror film laminated onto a specialized membrane structure—the highest standard for architectural mirror membranes worldwide.

1.2 Membrane Structure Engineering

The mirror membrane is not a simple flat mirror. Because it must deform under robotic actuation, the membrane must achieve both flexibility and reflective precision. Conventional glass mirrors cannot deform; metal-deposited films lack sufficient reflectivity. The solution adopted for null² is a three-layer structure: aluminum vapor deposition on a PET substrate, protected by an ETFE film overlay.

Tension management is critical. Excessive tension impedes deformation recovery; insufficient tension causes sagging under self-weight. null² sets pre-tension values at 1.5 kN/m (warp direction) and 1.0 kN/m (weft direction), derived through iterative numerical simulation and physical testing.

The 300,000-cycle durability test assumes approximately 1,630 deformations per day over the 184-day Expo period. Actual daily deformation frequency is projected at 500–800 cycles, providing a safety factor of 3–4x. Testing includes JIS A 1414-compliant accelerated testing and environmental chamber testing simulating Osaka summer conditions (35C, 80% humidity).

1.3 Reflection Characteristics and Distortion Control

The balance between accurate flat-mirror reflection and intentional deformation-induced distortion is key to null²'s visual experience. In the undeformed state, the membrane provides accurate self-images; during deformation, it continuously distorts images to create the experience of "dissolution and reconstitution of the self."

Reflection image distortion is controlled as a function of membrane curvature radius and observer distance. For a convex mirror of curvature radius R [m] observed from distance d [m], image magnification approximates to M = R / (R + 2d). null² continuously controls R between 2m and infinity (flat), yielding magnification in the range 0.3–1.0.

2. Robotic Actuation

2.1 System Configuration

The robotic actuation system that dynamically deforms the mirror membrane is the core interactivity technology of null². Linear actuators positioned behind the membrane push and pull it based on commands from the digital twin, controlling the mirror surface geometry in real-time.

2.2 Control Architecture

The 640 actuators are divided into 8 zones for control purposes. Each zone contains 80 actuators and a local controller (PLC), operating based on commands from the master controller. Local controllers communicate via EtherCAT, achieving 1ms cycle time for high-speed synchronized control.

The master controller interfaces directly with the digital twin system, converting spatial simulation results into position commands for each zone's actuators. The conversion process employs real-time surface fitting algorithms, reconstructing smooth mirror surfaces from 256 (ceiling) or 384 (wall) discrete actuator positions.

Safety systems are built on independent safety PLCs (SIL 3 compliant). When visitors approach the membrane too closely, actuators automatically retract. When membrane tension exceeds allowable values, motion stops immediately. Each actuator includes an individual torque limiter for mechanical overload protection.

2.3 Motion Patterns and Wave Generation

Membrane deformation patterns are designed not as simple convex/concave shapes but as "waves." By converting 2D wave equation simulations into actuator positions, the system generates ripple-like motions across the membrane surface—concentric circles like stones dropped in water, or interference patterns from multiple wave sources.

When the sensor network detects a visitor's position, that position is fed into the wave simulation as a "wave source." When multiple visitors are present, each becomes a wave source, and their waves interfere with each other. This "wave interference" functions as a physical analogy for the Buddhist concept of mutual interpenetration.

3. Digital Twin

3.1 Real-Time Spatial Reconstruction

null²'s digital twin reconstructs the pavilion's physical space in real-time as a virtual environment. It is not merely a 3D model but a living digital copy that synchronizes visitor positions, membrane deformation states, acoustic environments, and lighting conditions in real-time.

3.2 Physics Simulation Engine

At the core of the digital twin is a physics simulation engine that integrates FEM-based membrane deformation simulation with ray-tracing-based optical simulation to compute in real-time "how this membrane shape will appear to visitors."

The membrane FEM model consists of approximately 10,000 triangular elements, with stress, strain, and displacement analyzed in real-time for each element. GPU-parallel computation derives the entire membrane shape from 640 actuator positions within 1ms.

The optical simulation calculates reflection directions from the normal vector at each membrane point, predicting mirror images from the visitor's viewpoint. These predictions determine actuator control targets: an inverse-rendering approach that optimizes membrane shape backward from the visual goal of "how we want it to look to the visitor."

3.3 Adaptive Control and AI Feedback

The digital twin incorporates a machine-learning-based adaptive control module. By accumulating visitor data throughout the exhibition period—attendance by time slot, movement patterns, dwell times—the system optimizes membrane deformation patterns. Early in the Expo, the system runs pre-programmed base patterns; as data accumulates, it evolves toward patterns that enrich the visitor experience. This is the technical foundation for null² being a "living pavilion."

4. Sensing Network

4.1 Multi-Modal Sensing Configuration

null²'s sensor network is a multi-modal system for high-precision detection of visitor presence and movement. Multiple sensing modalities are combined to achieve stable detection under the optically challenging conditions of a mirror-enclosed environment.

4.2 Sensing Challenges in Mirror Environments

Sensing within a mirror-enclosed space presents unique difficulties. LiDAR infrared lasers specularly reflect off mirror surfaces, potentially returning incorrect distance values. Camera-based systems cannot distinguish mirror images from real objects.

For LiDAR, infrared-absorbing markers (940nm absorption paint, transparent in visible light) are applied in a 10cm grid on the membrane surface, enabling clear membrane identification. This separates membrane reflection signals from visitor reflection signals.

Floor pressure sensors, unaffected by optical conditions, serve as the most reliable visitor detection method. Deep-learning models estimate visitor count, position, and direction from pressure distributions.

Microphone arrays perform position estimation through beamforming of visitor voices and footsteps. While sound also reflects off mirror surfaces, Direction of Arrival (DOA) algorithms leveraging time differences between direct and reflected sound achieve high-precision source localization.

4.3 Sensor Fusion

Data from all sensors is integrated via an Extended Kalman Filter (EKF) based sensor fusion algorithm. Each sensor's uncertainty is modeled, and state estimation weights data from higher-confidence sensors. Fusion output includes tracking ID, 3D position (x, y, z), velocity vector, and estimated posture (standing/seated/wheelchair) for each visitor.

5. Subwoofer Vibration System

5.1 Acoustic Design for Tactile Experience

null²'s subwoofer vibration system is not an audio system for "listening"—it is a tactile system for "feeling." Sub-20Hz infrasound is transmitted through the floor to visitors' bodies, from the soles of the feet through the entire body, creating a physical experience of emptiness (ku).

5.2 The Physicality of Infrasound

Human hearing generally functions in the 20Hz–20kHz range, but sub-20Hz infrasound is not inaudible; rather, it is not perceived as "sound." The body detects it as vibration. Near the thoracic cavity resonance frequency (~10Hz) and abdominal resonance (~8Hz), powerful tactile experiences are produced through the viscera.

null²'s sound design draws on the reverberation characteristics of Japanese Buddhist temple bells (bonsho). A bonsho's resonance has a fundamental frequency around 64Hz, but its sub-harmonics (1/2 and 1/4 harmonics) reach into the 16Hz and 8Hz infrasound range. The experience of feeling a temple bell's vibration through the entire body is something many visitors to Japanese temples will recognize. null² recreates and extends this experience through technology.

5.3 Spatial Resolution and Wavefront Control

The 24 subwoofers are arranged in a grid pattern on the floor, with amplitude and phase independently controlled for each unit. This enables diverse vibration patterns: focusing vibration at specific locations, uniform whole-space vibration, or ripple-like propagation. Using visitor position data from the sensor network, vibration focal points can be formed beneath individual visitors' feet and made to follow their movement, creating the sensation that "the space is responding to me."

6. Integrated Control and Fail-Safe Design

6.1 Three-Layer Control Architecture

• Layer 1 (Experience Control): Digital twin generates target spatial states. Includes adaptive control based on visitor data. 60Hz update rate.
• Layer 2 (Device Control): Local controllers for each subsystem (actuators, subwoofers). Executes tracking control to target values. 1000Hz update rate.
• Layer 3 (Safety Control): Independent safety PLC. Monitors all systems for anomalies and executes emergency stops. SIL 3 compliant.

6.2 Fail-Safe Design

As an Expo pavilion requiring 184 consecutive days of operation, fault tolerance is paramount.

• Single actuator failure: Adjacent actuators compensate. Visual impact is localized; overall experience unaffected.
• Zone controller failure: Automatic failover to spare controller (N+1 redundancy). Switchover under 500ms.
• Digital twin failure: Automatic transition to preset pattern mode. Actuators continue on pre-programmed wave patterns.
• Sensor network failure: Degraded mode. Visitor-tracking features stop but preset patterns continue.
• Power loss: UPS executes safe shutdown sequence. Actuators return to neutral position.
• Earthquake detection: Accelerometer (intensity 3+) triggers full actuator stop. Membrane held at neutral tension.

6.3 Thermal Management

Osaka summers reach 35C and 80% humidity. The mirror-enclosed space risks greenhouse-effect temperature rise. null² implements ventilation openings in the membrane structure for natural convection, under-floor HVAC ducts (200kW cooling capacity), automatic shading sequences when membrane temperature exceeds 60C, and WBGT-index-based climate control for visitor comfort.

7. Construction and Maintenance

7.1 Design for Disassembly

null² is designed from the outset for post-Expo disassembly and recycling (Design for Disassembly: DfD). All structural joints use bolted connections; welding is minimized. 98% of steel is recyclable, with planned reuse at GREEN×EXPO 2027 Yokohama. Mirror membranes are designed as consumables, with two planned replacements during the Expo period. Each zone can be replaced independently, allowing continued operation during maintenance. Replacement time is approximately 4 hours per zone (overnight).

7.2 Predictive Maintenance

Software updates to the digital twin and sensor network are deployed OTA during overnight closure periods. Actuator operation logs are analyzed by machine learning models for failure prediction, enabling predictive maintenance that catches problems before they affect the visitor experience.