Authors: Harsh Sharma, Dr. Akshya Bhardwaj
Abstract: Delivering responsive augmented reality (AR) experiences on mobile devices requires maintaining high rendering performance while operating within the constraints of limited processing power, memory, and battery resources. This study introduces a benchmarking framework developed with Flutter that evaluates the runtime performance of AR applications using Google ARCore on a physical Android device. The proposed framework continuously records important runtime metrics, including frame rate, frame time, CPU utilisation, memory consumption, estimated power usage, and motion-to-photon (MTP) latency during live AR sessions. Six optimisation techniques—distance-based level of detail, visibility control for distant objects, adaptive rendering quality, lazy asset loading, background-isolate data export, and batched platform metric collection—can be enabled independently to examine their individual and combined impact through controlled A/B experiments. Performance evaluation was conducted on a Realme RMX5070 smartphone running Android 16 under identical environmental conditions using a Release build. Experimental results demonstrate that enabling all optimisation strategies increased the average frame rate from 92.94 FPS to 103.27 FPS, representing an improvement of approximately 11.1%. Average memory usage decreased from 286.5 MB to 267.7 MB, while mean UI frame time was reduced by 9.4%. Among the evaluated techniques, adaptive quality adjustment produced the largest individual improvement in rendering performance. CPU utilisation remained consistently close to full capacity throughout all experiments, whereas GPU utilisation metrics were unavailable because of platform restrictions. The proposed benchmarking framework offers a reproducible methodology for analysing Flutter-based mobile AR applications and provides a practical foundation for future optimisation research across different devices and AR platforms.
