JoWUA

Abstract

Adaptive AR/VR interfaces are crucial for wearable computing, aiming to deliver a context-aware and personalized user experience. Existing adaptation approaches based on static policies or minimal contextual knowledge often fall short in dynamically adapting to ever-changing cognitive and behavioral patterns. This paper presents a Self-Evolving Neural Policy Algorithm (SENPA) for adaptive AR/VR interfaces, which enables continual optimization of interaction strategies based on multi-modal user data in real-time. A wearable system incorporating eye tracking, motion sensing, physiological sensors, and contextual data was developed to model the cognitive-behavioral state. The proposed SENPA system integrates neural policy learning and an evolutionary optimization algorithm for continuous policy improvement and personalized interface adaptation. The performances of SENPA and DQN, PPO, SAC, and A3C learning approaches were evaluated in a wearable AR/VR experimental environment. Experimental results showed that SENPA achieved the best performance in attention retention rate (93.8%), task completion rate (97.2%), rendering performance (94 FPS), and network throughput (101 Mbps). According to one-way ANOVA statistical tests, there are significant differences (all p < 0.001) in the four learning approaches on these four major performance metrics. The ablation study demonstrates the significant impact of the cognitive modeling, behavioral analysis, evolutionary learning, and adaptive rendering components of SENPA. This suggests that SENPA improves cognitive engagement, behavioral interaction, and system reactivity effectively in a wearable AR/VR context and shows great potential for future work.