Global e-commerce giant Amazon has announced that it has deployed up to 1 million robots within its logistics network, a figure that is approaching the size of Amazon's warehouse workforce. At the same time, Amazon has also launched an artificial intelligence system called DeepFleet, which optimizes the movement paths of the robots, successfully enhancing operational efficiency by 10% and significantly reducing delivery costs.
Currently, Amazon employs around 1.5 million people globally, with approximately 1.2 million of them working in warehouses. The company has deployed robots across more than 300 logistics sites, with 75% of its distribution centers utilizing robotic assistance. Reports indicate that last year, the average number of employees per facility dropped to 670, marking the lowest figure in sixteen years. As automation advances, the number of packages managed by each employee has increased nearly 22 times over the past decade, rising from 175 to close to 3,870, showcasing the efficiency gains brought by automation.
DeepFleet, as a generative AI model, harnesses Amazon's internal logistics data along with AWS tools like Amazon SageMaker to orchestrate the movements of robots throughout the fulfillment network, much like a traffic control system. By mimicking smart traffic management concepts, this system significantly reduces congestion within warehouses and optimizes routes, enhancing the efficiency of customer order processing.
Amazon stated that DeepFleet can better position products closer to customers for faster delivery and cost reduction. This AI model continuously learns and improves, discovering new optimal methods for robotic collaboration, thereby enhancing work efficiency and safety while creating new job positions in maintenance and technical operations. Since 2019, over 700,000 employees have participated in training for related roles.
Amazon's robot deployment began in 2012 with the introduction of warehouse shelf-moving robots, and it currently operates 11 different types of robots. Among them, the flagship robot, Hercules, is responsible for moving entire shelf units closer to workers, minimizing walking time and speeding up item selection. Meanwhile, Pegasus is equipped with wheeled units to efficiently transport packages between fulfillment centers.
Additionally, Amazon's first fully autonomous mobile robot, Proteus, can now safely collaborate with humans in warehouses, while the tactile robot Vulcan uses force sensors and AI technology to assist in tactile operations, alleviating the repetitive workload for workers. The bipedal robot Digit, developed by Agility Robotics, is currently undergoing testing and is expected to take on responsibilities such as unloading trucks in the future.
In the robotic system Sequoia, these mobile robots, gantry systems, robotic arms, and ergonomic workstations are effectively integrated to achieve containerized management of inventory. The robotic arm Robin uses vision and suction to sort and transfer packages between different systems; meanwhile, Cardinal is responsible for lifting heavy items and transporting them to delivery trucks, while Sparrow utilizes suction cups and computer vision to select individual items from tote bags.
The Titan, a more powerful version, is designed to handle heavier inventory, while the modular mobile robot platform Xanthus can adapt to various warehouse operation needs through interchangeable attachments. Forrester analyst Rowan Curran points out that DeepFleet showcases the growing maturity of generative AI suppliers in developing non-verbal models, highlighting the scalability value of these models and the numerous potential use cases.
Other companies in the industry, such as Netflix and Microsoft, are also actively developing foundational models for new uses. Netflix has rolled out recommendation models based on members' viewing history, while Microsoft is focusing on generative AI models tailored for gameplay. Experts believe that as long as companies have abundant data and robust training capabilities, they can create a variety of application cases from these foundational models.
