A single piece of reconfigurable cloth has proved capable of replacing multiple dedicated virtual reality props by adopting three distinct physical states—a flat two-dimensional surface, a folded three-dimensional shape, and an actively transforming configuration. Each state supported different types of VR interaction within the same portable interface, removing the need to carry separate props for each virtual task.
- Embedded sensing fibres captured touch and deformation signals without relying on external cameras, with multi-channel outputs used to identify the current state and action combination.
- The work drew on cross-disciplinary expertise spanning textile materials, interaction and configuration design, sensing and signal acquisition, and machine learning, with industry partner support for fabric design.
- By encoding a predefined folding sequence as a command input, the transforming state allowed mode switching or tool switching without swapping props or recalibrating alignment.
- The paper 'Origami-inspired fabric makes one cloth act as many VR controllers' was published in National Science Review, with researchers at Donghua University leading the work.
THE STUDY: The research has addressed a core technical challenge in using fabrics as VR interfaces: controllability. Although cloth softness and deformability are attractive for portability and comfort, the same target configuration can be reached via many different folding paths, making reliable state recognition difficult. The team has responded with topology-guided deformation, employing preset fold lines, connectivity rules, and rigid-soft partitioning to constrain the fabric's behaviour.
- The same target fold configuration can be achieved through many different paths, and the cloth can collapse, twist or relax during grabbing and manipulation, generating unstable tactile boundaries and ambiguous signals.
- The approach has channelled the fabric's vast continuous deformation space into a small set of discrete, predictable and repeatable configurations, converging the cloth's behaviour into stable, standardised states.
- The resulting configurations have yielded clearer face, edge and corner references after folding, reducing state ambiguity and enabling standardised interaction procedures that are easier to learn and reproduce.
- Fabrication cost, storage, transport burden and setup time compound the deployment challenge, becoming critical in constrained settings such as astronaut cabin or extravehicular operation training environments.
- Touch is harder to simulate than vision in VR; physical boundaries and tactile cues are what help stabilise user control and spatial perception during virtual interactions.
WHAT THE DATA SHOWS: A seven-channel FTHP prototype combined with a lightweight convolutional neural network decoder has demonstrated 92.4% action recognition accuracy across all three states, supporting multi-mode interaction recognition and VR control. In the flat state, it has served as a surface for menus, panels, sliders and continuous control; in the folded state, it has provided three-dimensional boundaries for clicking, sliding, grasping and rotating; in the transforming state, the act of folding itself has triggered mode changes.
- The foldable, low-burden and portable nature of the interface has made it promising for constrained-space operation training and human-machine interaction scenarios, the authors stated.
- The three-state framework has provided a scalable route for multi-task VR interaction, reducing recalibration demands and eliminating the need to swap physical props when switching between tasks.
- The authors have noted that the same design philosophy could inspire portable, scalable immersive interfaces for broader applications where easy deployment and reusability matter beyond VR training.
- In the folded state, the cloth was shaped into simple 3D geometric controllers such as cubes or prisms, providing more explicit boundaries for clicking, sliding, grasping and rotating.
- Multi-channel signals were decoded by a lightweight model to identify state and action together, using touch cues in flat mode, face-and-edge data in folded mode, and time-sequenced triggers in the transforming state.
