Deep Learning-Enabled Smart Serving Robots for Sustainable Supply Chain and Logistics Automation, A Review

Abstract

The rapid expansion of e-commerce and global supply chains has created an urgent need for efficient, sustainable, and automated logistics operations. Smart serving robots, including autonomous mobile robots, robotic arms, and drones, are increasingly deployed in warehouses and last-mile delivery to improve productivity and reduce human labor. Deep learning has become a critical enabler for these robots, providing advanced capabilities in perception, navigation, and decision-making within complex environments. This systematic review aims to map the existing literature on deep learning-enabled smart serving robots for supply chain and logistics automation, synthesize reported outcomes related to operational efficiency and sustainability, and identify key technological and operational challenges. We conducted a comprehensive literature search following the PRISMA guidelines across multiple academic databases and search engines. Studies were screened based on predefined inclusion and exclusion criteria, with data extraction and quality assessment performed independently by two reviewers. The synthesized evidence reveals a rapidly growing but fragmented research landscape. Deep learning techniques, particularly convolutional neural networks for object recognition and reinforcement learning for path planning, are widely applied in warehouse order picking, inventory management, and last-mile delivery. Operational outcomes consistently demonstrate improvements in path planning efficiency and task completion rates. However, sustainability outcomes are often addressed superficially, with few studies providing quantitative metrics on energy savings or lifecycle impacts. Key challenges include sensor noise, the sim-to-real gap, high deployment costs, and a lack of standardized sustainability benchmarks. We conclude that deep learning significantly enhances robot autonomy and adaptability in logistics, yet critical gaps remain in the substantive integration of sustainability metrics. Future research should prioritize standardized evaluation frameworks and explore multi-robot collaborative systems to advance both automation and sustainability.