Core idea of this paper:
– To evaluate the performance of edge devices for AI workloads
To assess the effectiveness of AI technologies
To provide a comprehensive hardware comparison
The challenge
The increasing demand for edge computing in QSR kiosks, fueled by the post-COVID shift to self-service technologies, highlights the need for this study. As AI workloads like ASR, NLP, and CV become integral to enhancing customer experiences, selecting small form factor hardware that can efficiently handle these tasks with low latency, high performance, and optimal thermal management is crucial in dynamic QSR environments.
Summary of Key Findings
The key findings of the benchmarking study reveal that a single Intel device, such as the i7-1165G7, can effectively handle basic QSR kiosk workloads, delivering a 2.2s audio response time and strong 3D avatar graphics performance. For intermediate profiles, a pure edge setup using Intel i5-1135G7 and i7-1165G7 devices outperformed hybrid and cloud configurations, offering faster response times and higher graphics performance. Hybrid and cloud setups showed medium usability with slower response times. In comprehensive profiles, the hybrid setup provided the best end-to-end response time of 4.1s, while the pure edge setup had longer response times. The study also highlights the importance of using OpenVINO optimized AI models for improved performance, especially in ASR workloads on Intel i5 devices.
Annexure
The study outlines the technology components that enable a seamless QSR ordering experience, including Computer Vision for vehicle and customer identification, Facial Detection for customer consent, Automated Speech Recognition for contactless ordering, Natural Language Processing to understand customer intent, and Digital Signage to display relevant content.
The ordering process begins with the kiosk displaying trending dishes on the home page. Upon customer arrival at the drive-through, the kiosk detects the vehicle and performs facial recognition. Interactions can be initiated by the kiosk’s greeting, customer voice command, or touchscreen buttons. Customers select menu items either by tapping the screen or using voice commands, with the option to view and modify the cart. The process concludes with a review page where customers finalize the order and select their payment method.
