Experience Paper: Adopting Activity Recognition in On-demand Food Delivery Business

2025-09-29

AI Summary

This study addresses practical challenges in on-demand food delivery—specifically, low accuracy in rider behavior recognition and poor model generalizability across diverse real-world conditions. We present the first large-scale deployment of Human Activity Recognition (HAR) on a nationwide instant-delivery platform. Leveraging our proprietary LIMU-BERT foundation model and fusing multi-source smartphone sensor data (e.g., accelerometer, gyroscope, GPS), we develop an adaptive HAR system robust to complex operational environments, accurately classifying core delivery activities—including cycling, walking, and stationary states. The system achieves the first nationwide HAR deployment in logistics, covering 367 cities and over 500,000 riders, and directly supports critical business functions: dynamic workforce dispatching, safety monitoring, and performance evaluation. Our approach significantly enhances recognition robustness and cross-domain transferability; we further release the pre-trained model publicly to advance standardized industrial adoption and community-driven development of HAR in real-world logistics scenarios.

Technology Category

• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Application Domains: Transportation
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

This paper presents the first nationwide deployment of human activity recognition (HAR) technology in the on-demand food delivery industry. We successfully adapted the state-of-the-art LIMU-BERT foundation model to the delivery platform. Spanning three phases over two years, the deployment progresses from a feasibility study in Yangzhou City to nationwide adoption involving 500,000 couriers across 367 cities in China. The adoption enables a series of downstream applications, and large-scale tests demonstrate its significant operational and economic benefits, showcasing the transformative potential of HAR technology in real-world applications. Additionally, we share lessons learned from this deployment and open-source our LIMU-BERT pretrained with millions of hours of sensor data.

Problem

• Deploying human activity recognition in food delivery
• Adapting LIMU-BERT model for nationwide delivery platform
• Demonstrating operational benefits through large-scale HAR implementation

Innovation

• Adapted LIMU-BERT foundation model for delivery
• Deployed human activity recognition nationwide
• Open-sourced pretrained model with sensor data