Intelligent wearable olfactory interface for latency-free mixed reality and fast olfactory enhancement

Olfaction, a crucial physiological system, significantly impacts human interaction with the environment, influencing hazard detection, social behavior, emotional regulation, and memory recall. Despite its importance, olfactory interface technology lags behind haptic feedback, suffering from limitations in size, response time, and power consumption. Current wearable olfactory interfaces, often based on arrays of odor generators (OGs), struggle with second-level response times, leading to noticeable latency. Metaverse applications, for example, demand immediate and immersive experiences, requiring minimal delay. Further challenges include miniaturizing the size, optimizing power consumption, and increasing the number of odor channels. Existing systems typically require bulky equipment and professional operation, limiting user accessibility. This research aims to overcome these limitations by developing a wearable olfactory interface with sub-second response time, centimeter-scale size, a high-density array of OGs, and micro-watt power consumption.

Existing literature highlights the need for improved olfactory interfaces. Several studies explore different OG mechanisms, such as piezoelectric atomizers and controllable phase change of odorous wax, but these methods suffer from response time limitations in the order of seconds. The lack of attention to olfactory interface development compared to haptic feedback technologies has led to long-term technical stagnation. Existing wearable solutions also face challenges in miniaturization, power efficiency, and channel density. The need for a compact, energy-efficient, and high-channel-count device that delivers rapid olfactory feedback is widely acknowledged in the field.

This study introduces a novel AI-driven, wireless olfactory interface using miniaturized OGs. The OGs utilize airflow and heat for odor generation, controlled by a mechanical actuator opening/closing breathing holes, and heating temperature adjusting odor concentration. A multilayer design features a PET layer with breathing holes, a magnet-actuated structure, a cotton odorant container, a gold-chromium heating platform, and a 3D-printed support. An external power management system provides voltage for the electromagnetic coil and heating electrode. A thermistor monitors temperature, enabling precise control. The system's mechanical stability is analyzed using finite element analysis (FEA). Experiments optimize breathing hole patterns, heating temperature, frequency, amplitude, and duty cycle of the electromagnetic coil to achieve ultra-fast response time and low-power consumption. The effects of airflow rate on odor concentration are also investigated. For MR applications, an AI algorithm predicts user motion to ensure latency-free olfactory feedback by considering factors such as wind speed, air temperature, pressure, user movement, and odor source location. Odor plume simulation uses a Gaussian plume dispersion model to account for diffusion and advection. The system integrates a flexible control panel and a positioning system for real-time tracking of odor sources and user location. Olfactory training is demonstrated using a 32-OG forearm-mounted interface with a flexible control panel, focusing on recognition rate and reaction time improvements through olfactory training protocols. A list-wise recommendation framework (LRF) based on reinforcement learning is introduced to personalize olfactory training plans.

The developed miniaturized OGs achieve a world-record response time of 70 ms, an array density of 0.75 units/cm², and latency-free operation. The OGs demonstrate excellent performance in terms of response time, power consumption (milliwatts), size (centimeters), stability, operating temperature range, and the number of odor options. Experiments show that the heating temperature directly influences odor concentration, while the frequency of the AC power has a negligible effect. However, the amplitude and duty cycle of the AC power significantly impact odor concentration. Increasing airflow suppresses odor concentration. The AI-driven MR system successfully delivers latency-free olfactory feedback, accurately predicting user movements and adjusting odor release accordingly. Olfactory training experiments with a 32-channel interface demonstrate significant improvements in recognition rate and reaction time for both with and without 1-hr pre-training. The LRF further enhances olfactory training, providing personalized plans that lead to faster improvement in olfactory capabilities. The system shows promise for aiding patients with anosmia in olfactory recovery.

The results demonstrate the successful development of a high-performance olfactory interface with unprecedented speed and miniaturization. The latency-free MR demonstration highlights the system's potential for immersive virtual experiences. The significant improvements observed in olfactory training, particularly with the personalized LRF, show the potential for clinical applications in olfactory disorders. The study's success in combining miniaturized hardware with sophisticated AI algorithms showcases a significant advancement in olfactory interface technology. This integrated approach has broad implications for various fields, bridging the gap between the digital and sensory worlds.

This research presents a significant advancement in olfactory interface technology, demonstrating a high-performance, wearable olfactory interface capable of latency-free mixed reality experiences and personalized olfactory training. Future work could focus on expanding the range of odors, improving the long-term stability and durability of the OGs, and further refining the AI algorithms for even more personalized and effective olfactory experiences. Exploring the integration with other sensory modalities and investigating applications in various clinical settings are also promising avenues for future research.

The current study primarily uses ethanol as an odorant, which may limit the generalizability to other odorants with different diffusion rates and properties. The odor plume simulation model is simplified and may not fully account for complex airflow dynamics in real-world environments. The LRF for personalized training is based on a limited number of volunteers and may require further validation with a larger and more diverse population. The long-term effects of the olfactory training remain to be investigated in larger clinical trials.