Enhancing VR Mandala Drawing and Natural Immersion for Attention Restoration with AI-Driven Bioadaptive Multimodal Interaction

Tiantian Geng; Huishan Lai; Lei Jing

doi:10.1145/3795011.3795053

Abstract

Restorative VR often relies on passive nature exposure. This work instead combines a calm natural surround with structured drawing and a physiological feedback loop that adapts during the experience.

Participants color a mandala inside a static 360° seascape. Heart-rate variability from a Polar H10 informs a lightweight machine-learning controller, which personalizes visual haze, ambient audio, and synchronized controller haptics.

In a within-subject pilot study with 11 participants, both conditions were experienced as generally restorative. The bioadaptive condition produced a distinct EEG regulation pattern and an exploratory association between perceived fascination and reaction-time improvement.

System

Bioadaptive restoration loop

The experience layer, physiological controller, and multimodal renderer form a closed loop around the user's changing autonomic state.

Closed-loop architecture connecting a Quest 2 nature and mandala experience, Polar H10 heart-rate sensing, AI-based arousal estimation, and visual, audio, and haptic rendering — ECG-derived R–R intervals are converted into HRV features and an arousal estimate. The resulting policy modulates fog, music, and controller haptics.

Study

Within-subject pilot

Eleven university students completed two counterbalanced sessions on separate days. Both sessions used the same nature-and-mandala experience; only the AI condition enabled bioadaptive multimodal feedback.

Each session included a six-minute pre-intervention Oddball task, a ten-minute VR intervention, a six-minute post-intervention Oddball task, and presence and perceived-restorativeness questionnaires. The analysis combined self-report, behavioral, HRV, and EEG measures.

Because this is a small exploratory pilot with a short intervention, the results describe condition-specific patterns rather than establishing broad clinical or long-term efficacy.

Results

EEG reveals a regulatory divergence

Both conditions received broadly positive subjective ratings, while the clearest between-condition distinction appeared in arousal-related Cz beta activity.

Pre and post EEG relative power and paired change scores for frontal theta, parietal alpha, and central beta under no-feedback and AI bioadaptive conditions — EEG relative power and paired change scores for the no-feedback (NF) and AI conditions. Dots represent individual participants.

Cz beta change differed between conditions in the pilot sample (Wilcoxon W = 9, p = .032, rank-biserial correlation = −.73): NF showed a positive median post–pre change, while AI showed a negative change. Frontal theta and parietal alpha differences were not statistically significant.

Exploratory analysis

Fascination and attention recovery

Perceived fascination was associated with faster post-intervention Oddball responses only when bioadaptive feedback was enabled.

Scatter plots relating perceived fascination to reaction-time improvement in no-feedback and AI bioadaptive conditions — Within-condition association between PRS Fascination and reaction-time improvement (N = 11): NF ρ = −.31, p = .362; AI ρ = .76, p = .007.

This condition-specific coupling is hypothesis-generating. Larger samples, repeated-use protocols, and factorial designs are needed to separate the contribution of multimodal stimulation from closed-loop adaptation.