Proximity3D: Shape from Capacitive Proximity on Sensing Manifold

AI summary

Overview

Research area: Computer vision and computational sensing — specifically shape reconstruction from non-visual, non-planar sensor measurements, with ties to robotic tactile/proximity sensing.

Technical level: Advanced. The work combines differential-geometry-flavored sensing representations (a curved "sensing manifold") with learned multi-view reconstruction, and assumes familiarity with shape-from-X reconstruction and capacitive sensing.

Scope (1 sentence): The paper proposes using a curved capacitive textile as a shape sensor and a feedforward multi-view model to reconstruct nearby object geometry from the capacitive proximity fields measured on that curved surface.

What This Paper Is About

Almost all shape reconstruction assumes measurements live on a flat sensing domain — an RGB image plane, a depth map, a planar array. This paper instead treats a curved capacitive textile as the sensor, so measurements live on a non-planar surface the authors call a sensing manifold. The goal is to recover the shape of an object near that surface from the capacitive proximity field the object induces, aggregating multiple known views of the sensor to do so.

Key Contributions

1. A curved capacitive textile as a shape sensor. The sensor surface itself is non-planar, so the measurement domain is a manifold rather than a planar grid.
2. A manifold-based measurement representation. Each scan is modeled as a capacitive proximity field defined on that curved manifold, arising from the interaction between the curved electrode layout and nearby object geometry.
3. A multi-view feedforward reconstruction model. The method aggregates these fields across known sensor views and outputs the observed object shape in a single feedforward pass.
4. Validation in both simulation and physical hardware. The authors report experiments with simulated and real capacitive proximity signals acquired on curved sensing surfaces.

Main Findings

• Reconstruction from capacitive proximity is feasible on curved surfaces: The abstract states that robust reconstruction was demonstrated from capacitive proximity signals acquired on curved sensing surfaces, in both simulated and physical experiments.
• Aggregating multiple views helps: The proposed model explicitly combines fields across known sensor views, indicating that a single view on the manifold is not the intended operating mode. The abstract does not state how many views are used or how performance scales with view count.
• The sensing manifold framing is central: Treating the curved electrode layout as a non-planar domain — rather than flattening it — is presented as the enabling idea, since the field is described as induced by the interaction between that curved layout and object geometry.
• A route to embodied near-field awareness: The authors position the result as pointing toward robotic near-field geometric awareness achieved through embodied sensing rather than through cameras. No quantitative accuracy, error, or comparison figures appear in the abstract.

Methodology in Plain English

The researchers replace the usual flat sensor — a camera chip or a planar electrode grid — with a piece of capacitive textile that is curved. Capacitive sensing detects nearby conductive or dielectric objects through changes in electrical capacitance, so when an object sits near the textile it perturbs the readings across the electrode layout. Because the textile is curved, those readings are not arranged on a flat grid; they are arranged on a curved surface, and the paper treats that surface as the domain on which a "proximity field" is defined.

To turn fields into shapes, the authors build a feedforward model — a learned network that maps input to output directly, without iterative optimization at inference. The model takes the proximity fields measured from several known viewpoints of the sensor and merges them, then produces the object's shape. The authors test this both with simulated capacitive data, where ground-truth geometry is available, and with a physical setup, where real textile readings are used.

Why This Matters

Impact on research: The paper challenges a default assumption in reconstruction — that measurements are defined over planar domains. If curved sensing surfaces can be modeled as manifolds and reconstructed from directly, that opens a family of sensing problems where the sensor's own geometry is part of the math rather than something to be corrected away. It also connects shape reconstruction to embodied sensing, where the sensor is part of a physical body interacting with its surroundings.

Real-world applications (as directions the work points toward, not demonstrated deployments):

• Robotic manipulation and grasping: near-field awareness of an object's geometry just before contact, useful when a gripper's surface is curved.
• Soft and conformable robotics: sensors that drape over non-flat robot skins, where a planar sensor array would not fit.
• Human-robot interaction and safety: detecting an approaching person or limb through a curved fabric covering.
• Wearable and smart-textile systems: textile surfaces that sense nearby objects or body geometry without cameras.

Industry relevance: Capacitive sensing is cheap, low-power, and already ubiquitous in touchscreens and consumer devices, and textiles are manufacturable at scale. A method that extracts 3D geometry from capacitive textile — rather than only detecting touch or proximity — could be attractive for robotics, automotive interiors, and consumer electronics, where cameras raise cost, power, privacy, or occlusion concerns. The abstract does not report cost, latency, or accuracy figures, so the practical readiness of the approach is not established by it.

Future Directions

1. Quantifying reconstruction accuracy and robustness. The abstract reports that experiments were run but gives no numbers; systematic error characterization, sensitivity to noise, and comparison against alternative sensing modalities remain to be shown.
2. Scaling to more complex or larger objects. It is unclear from the abstract how object size, complexity, or distance from the textile affect what can be recovered.
3. Closing the loop with robot control. The stated motivation is robotic near-field geometric awareness; the natural next step is using reconstructed shape online for grasping, avoidance, or manipulation rather than offline reconstruction.
4. Generalizing beyond capacitive textiles. Whether the sensing-manifold formulation transfers to other non-planar sensing modalities — other electrode materials, other field types, other curved geometries — is an open question.

Target Audience

Researchers and engineers working on computational sensing, shape reconstruction, tactile and proximity sensing, and robotic perception — particularly those interested in non-visual modalities and non-planar sensor geometries. It also suits practitioners in soft robotics and smart textiles who need geometric information from conformable surfaces, and graduate students looking for an example of reframing a sensing problem around the geometry of the sensor itself.

Authors’ abstract

Most shape reconstruction methods assume measurements defined over planar sensing domains, such as RGB images or depth maps. In this paper, we use a curved capacitive textile as a shape sensor, treating its surface as a non-planar sensing manifold. Each scan is represented as a capacitive proximity field on this manifold, induced by the interaction between the curved electrode layout and nearby object geometry. We introduce a multi-view feedforward reconstruction model that aggregates these fields across known sensor views and recovers the observed object shape. Simulated and physical experiments demonstrate robust reconstruction from capacitive proximity signals acquired on curved sensing surfaces, pointing toward a new route to robotic near-field geometric awareness via embodied sensing.