Artificial Intelligence-Enhanced Augmented Reality Training to Improve Mobility, Balance, and Prevent Falls in Older Adults: A Feasibility Study

Falls are a significant health concern among older adults and can contribute to injury, fear of falling, reduced physical activity, functional decline, and loss of independence. Exercise-based interventions that target balance, gait, strength, and functional mobility can reduce fall risk in older adults. However, implementation outside supervised clinical settings may be limited by insufficient training dose, lack of individualized progression, and difficulty maintaining engagement.

Augmented reality (AR) provides an opportunity to deliver balance and mobility training while allowing participants to interact with virtual training elements within their physical environment. Previous studies have investigated AR-assisted exercise and rehabilitation approaches in older adults, including balance, gait, and mobility training. Although these approaches have demonstrated potential benefits, further research is needed to determine how AR-based training can be individualized and implemented effectively in community-dwelling older adults. Incorporating machine learning-based adaptation may enable training difficulty to be adjusted according to individual performance and provide an appropriate and progressive level of challenge.

This study will evaluate the feasibility and preliminary effects of an artificial intelligence (AI)-enhanced AR balance and mobility training program in community-dwelling older adults aged 55-80 years. The study will use a single-arm, open-label, repeated-measures design. Participants will complete a baseline assessment (T1), followed by an 8-week AR training intervention and an immediate post-intervention assessment (T2).

The intervention will consist of 3-5 training sessions per week for 8 weeks, with each session lasting approximately 30 minutes. Participants will perform standing- and walking-based activities presented through wearable AR smart glasses. Training activities will include obstacle negotiation, path following, step targeting, turning, and directional-change tasks designed to challenge functional balance and mobility. Training sessions may be completed at McMaster University or, for eligible participants, in the participant's home. Before beginning home-based training, the home environment will be assessed for suitability, and the participant will complete an initial on-site training session under direct supervision.

The AR training application was developed by the McMaster University research team and runs on a dedicated smartphone connected to the AR glasses. The application incorporates a machine learning model that adapts task difficulty according to individual participant performance using a challenge-point framework designed to maintain an approximately 80% task success rate. When performance exceeds this target, the system iteratively increases task difficulty, for example by increasing target speed or decreasing target size. When performance falls below the target, the system reduces task difficulty, for example by decreasing target speed or increasing target size. The model uses positional and movement data collected through the AR glasses together with training performance metrics, including target success rate, accuracy, response time, and movement speed, to evaluate participant performance and inform adjustments to training difficulty.