Mitigating Viscoelastic Hysteresis in Soft Robotic Sensors

29 concepts24 papers1 proposal

When we interact with the world, our sense of touch feels instantaneous. If you press your finger against a table and lift it, your nerves register the release immediately. For soft robots and wearable electronics, however, replicating this immediate response is surprisingly difficult.

Soft sensors typically rely on elastomers—like silicone, rubber, or polyurethane—to provide the flexibility and compliance needed to mimic human skin or wrap around moving joints. But these materials are viscoelastic. Unlike an ideal spring that stretches and snaps back instantly, a viscoelastic material behaves partly like a liquid: its internal polymer chains slide past one another and take time to reorganize.

This physical lag introduces hysteresis and creep. When a soft sensor is deformed and then released, its electrical properties (such as resistance or capacitance) do not immediately return to their baseline. Instead, the sensor retains a "memory" of its previous state.

Ideal Sensor (Linear, No Hysteresis)        Real Soft Sensor (Hysteretic Loop)
       Signal                                      Signal
         ^                                           ^       / Loading
         |   /                                       |      /  __
         |  /                                        |     /  /  \
         | /                                         |    /  /    \ Unloading
         |/                                          |   /__/      \
         +-----------> Deformation                   +-----------> Deformation

As a result, a single electrical reading from the sensor does not map to a single physical state; the true state depends entirely on whether the material is currently being stretched, compressed, or released. As Ryan L. Truby notes in Distributed Proprioception of 3D Configuration in Soft, Sensorized Robots via Deep Learning, this non-monotonic relationship makes direct, instantaneous mapping between sensor readings and physical shape exceptionally challenging. Under a constant load, this manifests as signal drift, where the output continues to shift over time (Benjamin Shih).

During rapid or cyclic movements, this lag compounds into measurement errors, ghost signals, or "strain overshooting" (Chong Gao et al.). To build reliable soft robots and wearables, researchers attack this problem from two distinct angles: re-engineering the physical structure of the material to eliminate the memory effect, or training algorithms to understand and decode it.

Mechanical and Structural Countermeasures

One way to solve hysteresis is to prevent the bulk polymer chains from binding and lagging during deformation. By changing the physical architecture of the sensing medium, researchers can force the material to recover its shape more efficiently.

  • Micro-structuring and Porosity: Instead of using solid, flat films which trap energy and slow down recovery, sensors can be fabricated with microscopic or porous architectures. For instance, K.D. Wise developed a humidity sensor utilizing thousands of micron-scale polyimide columns in A High-Speed Capacitive Humidity Sensor to bypass traditional material memory effects. Similarly, Jing Shu et al. utilized kirigami-inspired porous structures to reduce structural constraints and minimize material lag during dynamic bending.
  • Alternative Substrates: Rather than relying on standard silicones, which are highly prone to creep, some systems swap out the base material entirely. In A Flexible and Robust Large Scale Capacitive Tactile System for Robots, Perla Maiolino et al. replaced traditional elastomers with industrial 3D fabrics, significantly reducing mechanical lag and eliminating temperature-induced ghost signals.
  • Sliding Mechanisms: To prevent the internal friction that causes hysteresis, materials can be designed to slide rather than stretch macroscopically. Fuhua Xue et al. introduced a "gradient stiffness sliding" (GSS) strategy in Ultra-sensitive, highly linear, and hysteresis-free strain sensors enabled by gradient stiffness sliding strategy, achieving near-zero hysteresis by decoupling the electrical contact changes from the viscoelastic recovery of the polymer matrix.

Algorithmic Memory for Physical Memory

If a material inherently possesses a memory of its past deformations, the software interpreting its signals must possess an equivalent memory to decode it. Because standard analytical models struggle with time-dependent nonlinearities (Xiaohan Zhu et al.), researchers turn to recurrent neural networks—specifically Long Short-Term Memory (LSTM) networks.

LSTMs are uniquely suited for this task because they maintain an internal hidden state that updates over time. This hidden state acts as a mathematical proxy for the physical state of the polymer chains. By training an LSTM on sequences of sensor readings, the network learns to differentiate whether a specific resistance value represents a loading phase or an unloading phase based on the preceding data points.

Physical System:  Deformation History  --->  Viscoelastic State  --->  Current Signal
                                                                             |
                                                                             v
Digital System:   Signal History       --->  LSTM Hidden State   --->  Predicted State

This approach has proven highly effective across several domains:

Sidestepping Material Limits with Vision

A third paradigm avoids embedded conductive networks entirely by using optical or vision-based sensing. Instead of measuring changes in electrical resistance through a bulk material, these sensors use an internal camera to observe the physical deformation of a soft gel or skin directly.

As Annan Zhang et al. demonstrate in Vision-Based Sensing for Electrically-Driven Soft Actuators, tracking physical boundaries or markers visually can sidestep the severe hysteresis and signal drift typical of embedded conductive tracks.

However, optical tracking does not completely escape the laws of polymer physics. The protective outer membranes or gels themselves are still elastomeric and subject to recovery delays. In their work on MiniTac, Wanlin Li et al. note that elastomeric membranes can still exhibit high levels of hysteresis, leaving residual signal artifacts during high-speed surgical palpation.

To resolve these remaining visual lags, deep neural networks are used to calibrate the visual data. Carmelo Sferrazza et al. implemented a transfer learning calibration layer to map deformation history to accurate force distributions, while Weiliang Xu 0006 et al. utilized deep neural networks to correct for the creep effect during rapid, continuous robotic manipulations. By pairing visual tracking with neural network calibration, the system can infer the true applied force, even when the underlying material is still slowly creeping back into place.

Key Foundational Work

A Variable Capacitance Touch Sensor (1977) — contributed 1 concept elements to this direction.
A High-Speed Capacitive Humidity Sensor (1998) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_sensor_hysteresis_979df0f5,
  title = {sensor hysteresis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sensor-hysteresis/rdr_979df0f5e3874ab39451f64c6effdf76},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_sensor_hysteresis_979df0f5_v1,
  title = {sensor hysteresis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sensor-hysteresis/rdr_979df0f5e3874ab39451f64c6effdf76/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake