Thermal & Vibrotactile Feedback

• Thermal and vibrotactile feedback are modalities that convert control signals into temperature changes and skin vibrations to recreate realistic tactile sensations.
• Engineering implementations use Peltier elements, inertial motors, and fluidic circuits, achieving metrics like 86.7% thermal estimation accuracy and sub-20 ms response times.
• Advanced rendering methods, including data-driven and impedance control approaches, enable multimodal haptic experiences in VR, telemanipulation, and assistive technology.

Thermal and vibrotactile feedback are the principal modalities for evoking thermal and mechanical sensations in artificial haptic systems, targeting applications from virtual and augmented reality to advanced human–machine interfaces and robot-assisted manipulation. These feedback modes are realized by converting control signals into time-varying temperature differences (thermal cues) or skin–surface vibrations (vibrotactile cues), enabling rendering of object material, texture, temperature, and compliance in increasingly rich and realistic forms. The rapid evolution of multimodal haptic hardware, actuation techniques, and data-driven rendering algorithms has catalyzed progress toward replicating the perceptual richness of human touch.

1. Engineering Principles and Device Architectures

Modern haptic devices combine distinct actuation methods to deliver thermal and vibrotactile feedback, often integrated in wearable or robot-facing form factors. Key examples include soft hydro-pneumatic rings (Cozcolluela et al., 23 Mar 2025), multi-actuator palm devices (Cabrera et al., 2022), flexible wearable arrays (Trinitatova et al., 2019), fingertip-scale electromagnetic actuators (Mun et al., 2024), and speaker-plus-Peltier configurations (Khan et al., 5 May 2025).

• Thermal Feedback: Actuation is achieved through thermoelectric (Peltier), resistive, hydraulic (fluidic), or convective (air–liquid exchange) means. Devices range from direct contact (metal heater, Peltier) to fluidic circuits for distributed heating/cooling (Cozcolluela et al., 23 Mar 2025).
• Vibrotactile Feedback: Vibrotactile cues are generated by inertial motors (eccentric rotating mass, ERM), voice-coil actuators, electromagnetic diaphragms, or solenoid-driven fluidic valves capable of both single-frequency and arbitrary waveform synthesis. Vibrotactile stimuli target frequencies up to several hundred Hz, mapping onto the perceptual range of Meissner’s and Pacinian corpuscles.

Some architectures integrate both modalities with additional kinesthetic feedback or force control, as in wearable palm devices with impedance control (Cabrera et al., 2022), fully soft combined rings (Cozcolluela et al., 23 Mar 2025), or palm and fingertip arrays for VR/AR (Trinitatova et al., 2019).

2. Governing Physical Models and Transduction Mechanisms

Quantitative rendering of thermal and vibrotactile stimuli requires explicit models mapping actuation to perceived skin sensation.

Thermal Models:

• For thermoelectric modules, heat flow follows the Seebeck effect: $$\dot{Q} = \alpha I - \frac{1}{2} I^2 R - k \Delta T$$, with $\alpha$ the Seebeck coefficient, $R$ internal resistance, $k$ thermal conductance, and $\Delta T$ the temperature differential (Khan et al., 5 May 2025).
• In fluidic–hydraulic actuation, the ring surface temperature $T_{\mathrm{display}}(t)$ is governed by skin–display contact resistance $R_{\mathrm{skin–display}}$, material thermal properties, and convective transfer equations, e.g.,

$$q''(t) = \frac{T_{\mathrm{skin}}(t) - T_{\mathrm{display}}(t)}{R_{\mathrm{skin–display}}} \,.$$

Solutions must account for thermal time constants ($\sim$4–6 s for accurate $0.5^\circ$C tracking in miniaturized hydraulic tubes (Cozcolluela et al., 23 Mar 2025)).

Vibrotactile Models:

• Vibrotactile output amplitude and frequency are mapped from actuation parameters (motor current, voltage, or solenoid valve state) to skin acceleration and displacement; for electromagnetic diaphragm actuators:

$F_L = N I \ell_{\mathrm{eff}} B_{\mathrm{gap}} \,,$

where $N$ is coil turns, $I$ current, $\ell_{\mathrm{eff}}$ effective coil length, and $B_{\mathrm{gap}}$ flux density (Mun et al., 2024).

• Fluidic vibrotaction is achieved by modulating pressure in a soft pouch at up to 300 Hz, with pressure transients directly correlated to sensation intensity (Cozcolluela et al., 23 Mar 2025).

Signal bandwidth and temporal performance are critical; sub-20 ms loop times are crucial for perceptual immediacy and correct stimulation (Khan et al., 5 May 2025).

3. Rendering and Control Methodologies

There are two predominant paradigms:

• Prescribed Physical Models: The actuation signal is computed as a function of target stimulus parameters (amplitude, frequency for vibration; temperature, gradient for thermal), which are set according to virtual object properties or task cues. For impedance-controlled palm displays, continuous-time dynamics

$$M_d \ddot{x}_y + D_d \dot{x}_y + K_d x_y = F_{\mathrm{ext}}(t)$$

define the force–displacement relationship rendered during interaction (Cabrera et al., 2022).

• Action-Based Data-Driven Rendering: Here, user exploratory action (e.g., press–lift, static contact, sliding) gates which modality is rendered (softness, thermal, or roughness, respectively). For the hydro-pneumatic ring, a press–hold–lift sequence triggers a mapped pressure rise, while sliding modulates valve switching for roughness cues via surface profile encoding (Cozcolluela et al., 23 Mar 2025).

Synchronization and closed-loop feedback are handled at 30–120 Hz via microcontrollers for actuation, with additional PID regulation for temperature (Khan et al., 5 May 2025). Vibrotactile patterns are synthesized either as prerecorded audio/haptic files or via real-time digital synthesis with amplitude/frequency modulation (Khan et al., 5 May 2025, Trinitatova et al., 2019).

4. Representative Implementations

Device/Prototype	Thermal Modality	Vibrotactile Modality	Integration Details
HapticVLM (Khan et al., 5 May 2025)	Peltier TEC (20x20 mm, ±2A)	Speaker (1–1000 Hz)	Synchronized via controller
Soft Ring (Cozcolluela et al., 23 Mar 2025)	Hydraulic (42.5°C/5°C water)	Pneumatic, 300 Hz valve	Data-driven actuation
STEM Actuator (Mun et al., 2024)	Resistive/EM heating (secondary)	Voice-coil + soft PDMS	PWM + current profiling
TouchVR (Trinitatova et al., 2019)	(Not thermal)	Coin motors + DeltaTouch	3D force vector synthesis
LinkGlide-S (Cabrera et al., 2022)	(Hardware ready)	(Hardware ready)	Palm impedance control

In HapticVLM, vision-language reasoning guides the selection of vibrotactile and thermal feedback patterns in real time, achieving $84.67\%$ haptic pattern recognition and $86.7\%$ thermal estimation (within $8^\circ$C) (Khan et al., 5 May 2025). The soft hydro-pneumatic ring allows action-dependent textures, with participants matching rendered to real textures at up to 90\% accuracy (Cozcolluela et al., 23 Mar 2025). The STEM electromagnetic actuator achieves 0.33 N indentation and 58 G peak acceleration at 210 Hz, yielding multimodal output in VR (Mun et al., 2024). Wearable palm devices integrate vibro-actuators or leave hardware stubs for future thermal upgrades (Trinitatova et al., 2019, Cabrera et al., 2022).

5. Psychophysical Performance and User Studies

Evaluation focuses on recognition accuracy, realism, discrimination capability, and temporal performance.

• Recognition Rates: For palm displays with pattern rendering, static pattern recognition rates reach $78.1\%$ overall ($\geq80\%$ for most patterns), and softness discrimination via impedance control achieves $84.4\%$ correct (Cabrera et al., 2022).
• Texture and Material Matching: Soft ring studies report up to 90\% accuracy for certain textures, with significant deviations attributed to limits in kinesthetic force rendering (softness) and roughness actuation amplitude (Cozcolluela et al., 23 Mar 2025).
• Thermal Perception: Peltier-based cutaneous cues driven by VLM-estimated ambient conditions yield $86.7\%$ correct classification within $\pm 8^\circ$C (Khan et al., 5 May 2025).
• Response Latency: Best systems achieve vibration latencies $<$5 ms (actuator) and full thermal transitions within 4–6 s, with overall system latencies kept under 20–60 ms for haptics (Khan et al., 5 May 2025, Cozcolluela et al., 23 Mar 2025, Mun et al., 2024).

Adjective ratings, user preferences, and comparative studies with vision- or force-based cues clarify the subjective strengths and persistent limitations of thermal and vibrotactile feedback compared to full kinesthetic or static force outputs.

6. Applications, Constraints, and Future Directions

Applications include:

• XR/VR immersion with cutaneous–thermal realism (Cozcolluela et al., 23 Mar 2025, Trinitatova et al., 2019),
• Telemanipulation and telepresence (Cabrera et al., 2022),
• Assistive technologies and accessible interfaces,
• Context-aware object/material identification (Khan et al., 5 May 2025).

Main constraints arise from system volume (e.g., external pumps in fluidics), actuator power and heat dissipation (thermal/Peltier elements), achievable force/thermal amplitude, and multi-modality integration bandwidth. Limitations in kinesthetic feedback reduce softness realism with current ring or fingertip devices, while fluidic circuits often lack rapid-switching or portability (Cozcolluela et al., 23 Mar 2025).

Future work targets:

• On-device miniaturization (e.g., micro-blowers for pneumatics, soft Peltier elements (Cozcolluela et al., 23 Mar 2025)),
• Closed-loop adaptation from multimodal sensing (e.g., temperature, slip, force feedback in VR (Cabrera et al., 2022, Mun et al., 2024)),
• Richer pattern libraries via generative models for real-time, user-adaptive rendering (Khan et al., 5 May 2025),
• Integration with vision-language or foundation tactile models for context-dependent haptic feedback (Khan et al., 5 May 2025, Yang et al., 2024),
• Cross-modal rendering for full-hand and multi-site coverage, and multi-contact, kinesthetic-thermal-vibrotactile blending.

7. Design Recommendations and Technical Guidelines

• Thermal:
  • Prefer direct-contact Peltier or hydraulic transfer for precise temporal control; ensure adequate heat sinking and PID loop stability (Khan et al., 5 May 2025).
  • For wearable integration, prioritize small form factors and safe temperature limits to skin (<45°C sustained, transient 55–65°C) (Cozcolluela et al., 23 Mar 2025).
• Vibrotactile:
  • Achieve $>$100 Hz bandwidth for surface and slip cues; consider voice-coil or solenoid fluidic valves for arbitrary waveform synthesis (Mun et al., 2024, Cozcolluela et al., 23 Mar 2025).
  • Coin motors and ERMs offer compact form but narrowband control.
• Control/Rendering:
  • Structure rendering algorithms to gate feedback modalities by user-exploratory action (sliding—roughness, static—thermal, pressing—softness).
  • For VR/teleoperation, implement impedance or admittance control to synchronize rendered force/thermal cues to environment dynamics (Cabrera et al., 2022).
• Evaluation:
  • Include both objective (accuracy, latency, bandwidth) and subjective (likert ratings, preference) metrics. Compare patterns against real-world references and adapt actuation waveforms for perceptual tuning (Cozcolluela et al., 23 Mar 2025).

Thermal and vibrotactile feedback are now established as practical and perceptually potent modalities for multimodal haptic interaction. Their fusion in soft, wearable, and data-driven interfaces continues to extend both the fidelity and diversity of artificial touch systems across virtual, teleoperated, and assistive domains (Cozcolluela et al., 23 Mar 2025, Khan et al., 5 May 2025, Cabrera et al., 2022, Mun et al., 2024).