Sensing Physical Twins

• Sensing physical twins are digital surrogates that accurately capture an object's external morphology and internal sensor layout using automated design and 3D printing.
• They leverage soft, liquid metal-based sensors that detect sub-Newton deformations, providing distributed, high-fidelity force feedback.
• Applications in robotic coral conservation and delicate manipulation showcase their potential for ethical experimentation and reinforcement learning.

Sensing physical twins refers to the technical processes, architectures, and devices by which the detailed physical characteristics, states, and behaviors of real-world objects are captured and rendered for analysis, control, or policy learning—usually via digital surrogates, or “digital twins.” This paradigm encompasses both the design of physical artifacts to enhance sensing (e.g., sensorized replicas or explicit labeling) and the computational approaches for integrating, fusing, and replicating sensor data in virtual environments, with recent work extending these concepts to object-centric, tactile, and distributed multi-modal settings.

1. Automated Design of Soft Sensorized Physical Twins

A state-of-the-art workflow enables the transformation of complex 3D scanned objects into soft, sensorized physical twins. Starting from a polygonal mesh, the geometry is voxelized into a binary 3D grid. Users define keypoints (e.g., the base and branch of a coral), which the system connects by tracing a non-intersecting path using 3D Dijkstra’s algorithm. This automates internal channel routing through intricate geometries while minimizing manual input and ensuring channels remain printable and free of topologically problematic features (such as loops) (Liow et al., 22 Sep 2025).

Step	Method/Algorithm	Output
Mesh acquisition	3D scan + voxelization	Binary 3D voxel grid
Path routing	Dijkstra’s shortest path in 3D	Non-intersecting internal channel
Fabrication	3D printing + post-processing	Soft body with embedded channel

This design method allows for the generation of application-specific physical twins that closely reproduce both the external morphology and interior topology of natural specimens (e.g., corals), providing customizability and high-fidelity mapping from biology to artifact.

2. Soft Liquid Metal Sensor Integration and Signal Properties

The core sensing mechanism is realized by infusing the fabricated channel with low-melting-point liquid metal (Galinstan), forming a soft strain sensor inherently compliant with the host structure’s geometry. The sensor’s principle of operation is resistance-based: mechanical deformation alters the channel’s geometry, increasing resistance in proportion to strain. Under controlled constant current, measured resistance changes are used to quantify deformation:

$R_\mathrm{norm} = \frac{\Delta R}{\Delta R_0}$

where $R_\mathrm{norm}$ is the normalized resistance change, $\Delta R$ is the change under load, and $\Delta R_0$ is the baseline (rest) resistance. Sensitivity studies in controlled bench-top experiments show that such sensors detect deformations below 0.5 N, with a monotonic, approximately linear region up to several newtons (Liow et al., 22 Sep 2025). The design thus supports distributed, object-centric force sensing that outperforms traditional discrete fingertip sensors for delicate interactions.

3. Applications in Delicate Manipulation and Conservation Robotics

Sensing physical twins find direct application in tasks demanding gentle, nuanced interaction with fragile objects—particularly as “sensing corals,” which are high-fidelity replicas of natural coral branches. These physical twins are deployed as testbeds avoiding the ethical and practical constraints associated with experimenting on live organisms. Demonstrated applications include:

• Automated Coral Labelling: During the loop-and-tightening of a wire for specimen tagging, the change in channel resistance provides feedback reflective of over-tightening, enabling the robot to reverse operations to prevent damage.
• Robotic Coral Aquaculture: In underwater settings, robotic graspers equipped with sensing corals receive real-time, spatially distributed force feedback, stopping closing motions before exerting excessive local force.

Such setups facilitate skill acquisition for autonomous systems via reinforcement learning with realistic and safe surrogates, advancing both conservation goals and policy learning frameworks.

4. Experimental Validation: Sensitivity, Repeatability, and Environmental Robustness

Quantitative validation encompasses both laboratory and aquatic scenarios:

• Bench-Top Compression: Instron compression tests reveal reliable detection of sub-0.5 N interactions and a pronounced increase in normalized resistance with increasing compressive force (up to ~8 N).
• Precision and Drift: Multiple cycles (from 50 to 500) demonstrate high repeatability; a minor baseline drift, attributed to Joule heating and material fatigue, can be corrected algorithmically (e.g., high-pass filtering).
• Underwater Grasping: Submerged experiments with robotic manipulators confirm stable, distributed sensing even with uncontrolled, variable loading—where traditional fingertip force sensors proved inconsistent.

The data support the assertion that these soft, sensorized physical twins provide distributed and reliable feedback, crucial for delicate, real-world manipulation (Liow et al., 22 Sep 2025).

5. Expanded Modalities and Technical Implications

Compared to the canonical approach using point tactile or optical force sensors, the use of distributed, embedded liquid metal sensors in soft replicas provides notable benefits:

• Distributed Sensing: Sensing twins deliver a spatially continuous deformation profile, unlike localized readings from fingertip sensors.
• Morphological Fidelity: By matching the morphology and mechanical compliance of the original sample, the device deforms and responds in a representative manner.
• Ethical and Practical Training: These twins serve as proxies in reinforcement learning and autonomy research, supporting safe experimentation and data collection without risking damage to rare or fragile samples.

The integration of cost-effective, reconfigurable sensor fabrication with automated path planning paves the way for rapid deployment in customized experimental or field-based manipulation tasks.

6. Representative Mathematical Expressions and Algorithms

• Normalized Resistance:

$R_\mathrm{norm} = \frac{\Delta R}{\Delta R_0}$

• Path Routing Cost Function (Dijkstra):

$$f(i) = \text{cost}(i) + \min_{j \in \mathrm{Neighbors}(i)} f(j)$$

with penalization for revisiting voxels to avoid channel self-intersection.

• The design suggests that further algorithmic refinement (e.g., through machine learning – driven structural optimization of channel routing) could enhance performance or adaptability.

7. Outlook and Broader Significance

Sensing physical twins, as realized through soft, 3D-printable, liquid metal–sensorized artifacts, establish a pathway toward more faithful, ethically viable, and experimentally robust platforms for manipulation and policy development in robotics and conservation. While initially demonstrated on coral analogues, this methodology generalizes to any application demanding high-fidelity, delicate, and spatially informed feedback, especially where the physical sample’s preservation is paramount. A plausible implication is that as these workflows mature, the coupling of automated geometric digestion, flexible channel routing, and distributed soft sensing will become central to both the science of biomimetic twin fabrication and the practice of embodied intelligence in robotics (Liow et al., 22 Sep 2025).