- The paper introduces CASAband, a fully untethered textile wristband with four independently controlled SMA actuator channels that produce 1.7 N force, 3.2 mm displacement, and a 1.34–6.59 Hz bandwidth.
- The system weighs 63 g including electronics and battery, runs for more than four hours, and achieved over 92% accuracy in identifying seven haptic patterns after minimal training.
- The paper demonstrates practical use in gesture-based collaboration and 1 km GPS navigation while highlighting actuator durability, thermal management, wiring integration, and closed-loop control as priorities for future work.
CASAband is a fully untethered, textile-based haptic wristband that integrates compliant amplified shape memory alloy (SMA) actuators to deliver spatial and temporal tactile feedback on the wrist. The device addresses a persistent trade-off in wearable haptics: electromagnetic and pneumatic systems offer strong mechanical performance but require heavy backends (often hundreds of grams), while thermal actuators are lightweight but historically limited to sub-0.33 Hz operation and largely binary cues. CASAband achieves 1.7 N blocked force, 3.2 mm free displacement, a −3 dB bandwidth of 1.34–6.59 Hz depending on actuation time, four independently controllable channels, silent operation, and a total system mass of 63 g including battery and electronics (2607.15533).
Design requirements and actuator selection
The design targets were derived from psychophysical literature rather than arbitrary specifications. The authors required over 1 N of force at 1.5 mm displacement, based on reported minimum detectable indentation on the forearm (1.5 mm) and contact pressure thresholds (~0.41 N), and prior devices achieving >80% localization accuracy at forces up to 0.6 N. A bandwidth of at least 1 Hz was justified by guidance and navigation tasks, which operate at task-level update rates of roughly 1 Hz and stimulate Merkel disks and Ruffini corpuscles sensitive to low-frequency stimuli. Actuator spacing was set at 35 mm (small wristband) and 40 mm (large wristband), consistent with two-point discrimination thresholds of 30.7–45.4 mm on the forearm.
The compliant amplified SMA actuator (CASA) builds on prior work by the authors (2607.15533). It uses two 200 μm GFRP compliant beams to amplify the 3–5% contraction of SMA wire into vertical displacement. Three design parameters were tuned experimentally: SMA wire diameter (0.05 mm selected for cooling time ~0.3 s), number of wire rows (six rows selected to meet the force target), and total wire length (148 mm, corresponding to a local maximum in mechanical work). A notable concession is that optimizing for force and displacement increased the actuator height by ~1.7 mm relative to the thinnest possible configuration; the authors judged this acceptable since the resulting profile remains below the ~1 cm thickness of comparable soft actuators.
Textile integration and untethered electronics
The wristband consists of two fabric layers bonded with heat-fusible film via a 2D lamination process requiring only laser cutting and heat pressing—no custom knitting equipment. Force transmission was maximized through a stiffness gradient: the outer fabric is considerably stiffer than the inner skin-contact layer, an approach analogous to distributed-stiffness knitting but achievable with off-the-shelf textiles.
Two fabrication-dependent variables materially affected output force: the width of the heat-bonded adhesive film and the donning preload. Characterization on an instrumented mock forearm (a 6-DOF F/T sensor with a 4 mm Ecoflex silicone layer) showed that a 5 mm adhesive width with preload between 0.25 N and 0.4 N maximized force; narrower adhesives caused buckling, wider ones insufficient stiffness transmission. Importantly, force output was similar across the 0.25–0.4 N preload range, so qualitative tension adjustment during donning suffices—a practical result for real-world use.
The onboard electronics (Arduino Nano 33 BLE, IMU, two 12 V step-up regulators, dual motor drivers, 7.4 V 350 mAh LiPo) weigh 49.2 g, bringing total device mass to 63 g. Under continuous cue generation at 10-second intervals, the device operated for over four hours per charge. This is among the lightest complete wrist-worn haptic systems reported when all operational components are counted; comparable untethered pneumatic or servo-based systems weigh 220–440 g.
Dynamic, durability, and thermal characterization
Bandwidth characterization revealed non-intuitive behavior. With 12 V inputs and actuation times from 25 ms to 100 ms, the −3 dB bandwidth ranged from 1.34 Hz (100 ms) to 6.59 Hz (25 ms). Displacement exhibited resonance-like peaks at specific frequencies (2 Hz for 50 ms, 5 Hz for 25 ms actuation), which the authors attribute to the interplay between cyclic heating and thermal recovery rather than mechanical resonance—an important distinction for anyone modeling such actuators. Embedding the actuator in the textile preserved low-frequency displacement (<2 Hz), likely because the stretchable fabric provides a bias force aiding shape recovery during cooling.
Durability testing showed 1127 cycles at maximum force before failure, with failure mode being SMA wire damage leading to non-uniform heating and snapping. The preload settled after ~300 cycles due to fabric movement, a detail relevant to long-term wear reliability. Thermal characterization showed skin-contact temperature converging to 31–32 °C under representative duty cycles (safe and within forearm thermal comfort), though under more aggressive conditions (1 s on at 8 V) temperature reached but did not exceed 45 °C—above the 40 °C threshold where continuous contact becomes potentially uncomfortable. The authors mitigated this by limiting pattern durations and relying on the longer inter-cue intervals typical of actual use cases (7–11 s).
Perceptual validation
A two-phase study with 10 participants evaluated localization and pattern identification. Participants localized single impulsive cues at four wrist locations with 97%, 96%, and 93% accuracy for high, medium, and low strength cues respectively. Nonparametric analysis (Friedman test with Bonferroni-corrected Wilcoxon post hoc comparisons) found a statistically significant difference only between high and low strength cues (p=0.0156 against a corrected threshold of p=0.0167); the low cue was deliberately set near the perceptual threshold, so inter-user variability in detection thresholds likely explains this gap.
Pattern identification covered seven cues varying in speed, direction, and magnitude (slow/fast clockwise and counterclockwise rotations, tap, heartbeat, squeeze). All exceeded 92% accuracy, with the highest confusion between tap and heartbeat patterns, attributed to similar principal frequencies. These results were obtained with minimal training (up to 7 minutes practice), supporting the claim that the cue vocabulary is intuitive.
Demonstrations
Two demonstrations established functional readiness beyond laboratory perception tasks. In a collaborative pick-and-place task, a Leader wearing a CASAband transmitted instructions via IMU-classified arm gestures (Madgwick-filtered quaternion orientation with threshold-based classification of six gestures) to a blindfolded, noise-isolated Follower. Over a 10-minute session, the Follower correctly identified 73 of 83 cues (88% accuracy)—notably lower than the controlled study's >90%, reflecting real-time gesture recognition errors—and still completed the target drawing because the Leader provided corrective feedback.
In an outdoor GPS navigation demonstration, a user followed a 1 km route with 16 waypoints using only wrist-worn directional cues delivered by a custom iOS/MapKit application, reaching the destination in 14 minutes at 4.3 km/h average walking speed—comparable to normal pedestrian pace. The cue protocol used an approaching cue (10 m before waypoint), a stop cue (squeeze within 5 m), and a thrice-repeated directional cue, a redundancy scheme worth noting for haptic navigation protocol design.
Limitations and open questions
The paper is candid about several constraints. Actuator lifetime of ~1100 cycles at maximum force is adequate for demonstrations but the authors acknowledge longer-term use across broader user populations will require improved durability. Wiring between actuators and electronics adds thickness and complicates fabrication; conductive textiles or stretchable electronics are proposed but not implemented. Control remains open-loop, using pulse and step inputs only—the nonlinear, hysteretic nature of SMA is not compensated by closed-loop force or displacement feedback, although prior work by the group suggests capacitive sensing integration is feasible. The gesture recognition pipeline relies on fixed thresholds rather than learned classifiers, and its 88% real-time accuracy leaves headroom. Finally, thermal management constrains sustained high-duty-cycle operation; whether the device can support denser cue schedules than those tested remains unverified.
Conclusion
CASAband demonstrates that thermally actuated haptic interfaces can simultaneously satisfy force, bandwidth, multi-channel spatial resolution, and untethered wearability requirements that have previously been met only individually. Its combination of >90% perceptual accuracy across seven spatiotemporal patterns, sub-65 g total mass, multi-hour battery life, and successful outdoor navigation constitutes a substantive advance for wrist-based haptic communication. The remaining obstacles—actuator fatigue life, integrated wiring, and closed-loop control—are well-defined engineering problems rather than fundamental barriers, making the platform a credible basis for subsequent work in hands-free guidance and sensory substitution.
