Adaptive Squeeze-Release Rhythms

• Adaptive squeeze-release rhythms are dynamic patterns that alternate between stress-induced squeeze phases and recovery-oriented release phases based on real-time physiological input.
• They integrate continuous sensing via PPG, LSTM-based state estimation, and tactile plus LLM-generated feedback to modulate rhythm in sync with user physiology.
• These systems enhance stress management and cognitive regulation by offering personalized, closed-loop biofeedback that outperforms static interventions.

Adaptive squeeze-release rhythms are temporally structured patterns in which biological or engineered systems alternate between phases of heightened activity (“squeeze”) and phases of relaxation or recovery (“release”), with the timing and nature of these phases dynamically modulated in response to internal physiological state or external contextual input. This concept, originally grounded in the analysis of musical and motor behavior, has recently been generalized to encompass a wide spectrum of adaptive, closed-loop rhythmic processes in human-machine interfaces, cognitive enhancement tools, and embodied biofeedback systems. Adaptive squeeze-release rhythms integrate continuous physiological sensing, algorithmic rhythm modulation, and real-time interactive feedback (including tactile and narrative cues), enabling personalized interventions that couple bodily action with psychological state regulation.

1. Underlying Principles and System Architecture

Adaptive squeeze-release rhythms are characterized by tightly coupled interaction loops between physiological measurement, real-time state estimation, and actuator-driven feedback. A paradigmatic system, such as NieNie (Yu et al., 20 Oct 2025), acquires physiological signals (heart rate and heart rate variability) through photoplethysmography (PPG) using a wearable such as an Apple Watch. Signals are analyzed in real-time using lightweight neural classifiers—specifically, a TensorFlow Lite–deployed LSTM trained on multivariate windows—with windowed features including electrodermal activity (EDA), temperature, and heart rate.

The classifier outputs a continuous stress probability $s(t)$, which directly governs the parameters of rhythm generation: during periods of high detected stress, the rhythm generator shortens its cycle period (accelerating squeeze-release transitions); as stress declines, the rhythm slows, promoting deeper, longer squeezes and releases optimized for parasympathetic activation.

Algorithmically, the squeeze-release cycle period is adapted according to:

$T(t) = T_0 + [1 - s(t)](T_{\text{max}} - T_0)$

where $T_0$ is the cycle period for maximal stress, $T_{\text{max}}$ is the period for a relaxed state, and $s(t)\in[0,1]$ encodes stress probability.

Tactile instructions and cues are delivered via a Bluetooth-connected soft pressure device, with synchronized visual or auditory signals rendered in a real-time engine such as Unity. This enables users to match their physical actions to an algorithmically adaptive rhythm, closing the physiological-behavioral loop.

2. Physiological Sensing, Signal Processing, and Feedback Adaptation

Physiological sensing is central to adaptive squeeze-release systems. Continuous collection of heart rate and HRV via wrist-worn PPG sensors, with data streaming through APIs (Apple HealthKit, CoreBluetooth), provides low-latency (<30 ms) inputs for rapid stress estimation.

Segmentation and analysis occurs in real time. Data are framed into sliding windows (e.g., 40 samples with step size 20), enabling the LSTM to predict $s(t)$ with high temporal resolution. Changes in $s(t)$ feed directly to the rhythm generator, which can promptly accelerate or decelerate the squeeze-release tempo to reflect acute elevations or reductions in arousal.

Rhythm modulation thus remains strictly state-contingent: if $s(t)$ spikes (indicating rising stress), the system transiently increases the pace of squeezes and releases. As $s(t)$ attenuates, the tempo is correspondingly slowed, facilitating a progression from arousal-matching to deep relaxation.

3. Real-Time Psychological Guidance Using LLMs

A distinguishing feature of recent implementations is the incorporation of LLMs to generate psychologically tailored feedback in real time (Yu et al., 20 Oct 2025). The LLM receives multimodal context—stress probability, adherence to prescribed rhythms (accuracy, slip events), and pressure profile characteristics. From these inputs, the LLM generates prompt narratives adapted to the user's interactive performance.

Examples of system-guided feedback include: “Keep up this steady rhythm; your body is syncing beautifully!” during periods of tight adherence, or “Try slowing down to find your rhythm again.” in response to rhythm deviation or user fatigue. Feedback is integrated seamlessly into the Unity-based environment, supporting a rich, multimodal experience.

Here, the LLM is not merely informational but instrumental in reinforcing the embodied loop: narrative direction, positive reinforcement, and strategy suggestions are co-evolved with the user’s physiological and behavioral state, surpassing the static scripting found in older stress-management apps.

4. Comparative Perspective: Traditional vs. Adaptive Interventions

Traditional stress management tools, such as mindfulness timers, fixed-script meditation guides, or static audio tracks, are generally open-loop and lack direct connection to ongoing physiology. These tools cannot dynamically sense user stress levels or responsively alter feedback.

Adaptive squeeze-release systems, by contrast, feature:

• Real-time, individualized physiological measurement and state estimation
• Rhythm generation and tactile actuation tightly coupled to current physiological state
• Dynamic, context-aware guidance rendered by LLMs, contingent on user performance

As a result, users interact within a closed, embodied feedback loop—modulating their own physiological state not through passive consumption but through continuous, actively guided tactile engagement. Initial user evaluations indicate that this approach can elicit more rapid and effective stress modulation than conventional apps.

5. Technical Implementation and Design Considerations

System modules are composed as follows:

• Physiological Interface: Apple Watch PPG sensor; data streamed over Bluetooth; custom preprocessing pipeline (windowing, normalization).
• On-Device Inference: Quantized LSTM, deployed via TensorFlow Lite; inference latency <30 ms; output: $s(t)$.
• Rhythm Control Algorithm: State-driven adjustment of cycle period using $T(t) = T_0 + [1-s(t)](T_{\text{max}} - T_0)$; on-the-fly update of squeeze-release instructions.
• Tactile Delivery and Sensing: Soft squeezable device with embedded pressure sensor (for physical input) and output actuators (vibration or pneumatic), managed via Unity and Bluetooth protocol.
• LLM-Driven Narrative: LLM (not specified; likely a transformer-based architecture) integrates real-time sensor data and interaction logs to generate prompts; output synchronized with visual and auditory channels in Unity.

All elements participate in a high-frequency closed feedback loop, with the stride of update matching user action timescale.

6. Broader Applications and Evaluations

The adaptive squeeze-release paradigm is generalizable beyond stress management. It encompasses synchronization therapy, tactile entrainment in sleep and relaxation contexts (see (Lee et al., 3 Jul 2025)), and embodied biofeedback for mental health and cognitive enhancement.

Empirical studies with NieNie (Yu et al., 20 Oct 2025) have shown that users experience improved engagement and more rapid downregulation of stress when compared to static-scripted applications. The continuous, tangible interaction—reinforced by responsive LLM feedback—supports both cognitive restructuring and sensorimotor regulation. Future research is oriented toward quantifying long-term efficacy, optimization of feedback modalities, and extension to new domains such as social interaction or motor rehabilitation.

7. Integration with Contemporary Rhythmic and Biofeedback Research

The adaptive squeeze-release rhythm model aligns with contemporary efforts to merge physiological sensing, algorithmic inference, and multimodal interactive feedback. It shares conceptual ground with earlier systems employing rhythmic haptic feedback for sleep and relaxation (Lee et al., 3 Jul 2025), but extends them by adding dynamic LLM-based narrative and embodied user engagement.

This integration realizes a fully immersive, adaptive, and embodied feedback architecture, mapping real-time physiological changes into interactively modulated rhythm patterns and dynamically updated linguistic guidance, thereby constituting a new standard in the design of bioadaptive stress regulation technologies.