Robotics and AI Evolution in the Chinese E-Commerce Fulfillment Market

Subscribe To Download This Insight

1Q 2018 | IN-5072

This is a preview of what is to come in the Chinese e-commerce fulfillment market. While warehouses in China still rely heavily on automated guided vehicles, the advancement in chipset, effector, and software technologies enables robots to mimic human workflow and work closely with workers, while being able to operate in unstructured environment.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


E-Commerce Fulfillment Needs More Automation


Since the emergence of Alibaba and, the e-commerce market in China has been growing at a rapid pace. This has led to an expansion in the logistics market, particularly in the segment of express delivery and contract logistics. Unlike Alibaba or, which rely mostly on in-house logistics arms, many major e-retailers rely heavily on third-party logistics service providers, such as SF Holdings, EMS (subsidiary of the China Post), and YTO Express.

According to the State Post Bureau, the express delivery market in China was estimated to be US$43.8 billion in 2015 and is expected to reach US$126.5 million in 2020. The prime example would be SF Holdings. Since its IPO in 2015, the leading Chinese company in third-party logistics has been growing by 22% year-on-year over the past 3 years.

However, compared to the customer-facing side of the business, e-commerce fulfillment processes remain labor-intensive and rely heavily on human employees. As of December 2017, Alibaba and still employ more than 2.5 million employees in warehousing and delivery, making them among the largest employers in the world. In comparison, Walmart, which has brick-and-mortar stores, employs 2.1 million staff. It is becoming increasingly costly to maintain such large workforces and China is facing a rapidly tightening labor market.

By deploying robots and AI solutions in three major aspects of e-commerce fulfillment, namely warehousing, order processing, and last-mile delivery, repetitive tasks, such as loading and unloading of goods, picking and sorting, and parcel delivery within the neighborhood, frees up valuable human resources to take care of more important tasks. Task automation not only improves service efficiency, but also brings service quality and accuracy, which is important for e-retailers, which rely heavily on service quality differentiation.

Goods-to-Person Systems Still Dominate in China


In the past few years, Alibaba,, Suning, and a few other Chinese e-retailers have been adopting goods-to-person systems in their warehouses. Robotics companies, such as Geek+ and Flashhold (a.k.a. Quicktron), are the greatest beneficiaries of this. For example, Geek+ has deployed more than 2,000 automated guided vehicles (AGVs) in Alibaba’s warehouse alone. Aside from the usual suspects, Flashhold has been working with the China Post, Vipshop, and BEST Logistics, deploying more than 3,000 AGVs across three markets.

When paired with intelligent warehouse management systems, the e-commerce fulfillment processes can be further optimized. Supported by PAI and PinoAI, their open-source machine learning platform, Alibaba and are able to perform intelligent interaction, semantic search, intelligent matching, and search strategy via their e-commerce portal. The customer insights collected from the platform would be used to train the goods-to-person system. The SKUs can now be sorted according to seasonal trends, and the positive and negative affinity between SKUs, allowing items to be picked and sorted in a speedier manner. A similar approach was commercialized by GreyOrange, another major goods-to-person system provider, with its GreyMatter solution.

Robots with Hands and Brains


It has been very clear from the beginning that a goods-to-person system has its limitations. Despite being very responsive and effective, the system requires high customization, long setup times, a structured environment with no human employee, and a greenfield environment. A goods-to-person system needs to be deployed in a warehouse equipped with fiducial markers on the floor to assist in navigation. This means the system is not suitable for brownfield warehouses, which can be very costly to retrofit and often feature mezzanine floors where goods-to-person systems cannot operate. A large segment of the contract logistics market is excluded from the robotics market. Understanding this limitation, Geek+ and Flashhold are introducing a premium tier of products equipped with simultaneous localization and mapping (SLAM) technology. Equipped with sensor fusion technology and extended Kalman filter, these autonomous mobile robots (AMRs) can operate in unstructured environments. This allows them to work closely with humans and they can be deployed in brownfield warehouse environments.

Moving beyond that, these AMRs need to have an effector system. Many existing AMRs serve predominantly as trolleys for goods. AMR vendors, such as inVia Robotics, IAM Robotics, and Magazino, have either deployed vacuum or mechanical grippers to retrieve items from storage racks. One key component of the effector system is machine vision. In the past, the adoption of machine vision components was slow, due to their high cost. The cost has slowly declined over the years, thanks to the emergence of low-cost image processing units with the capabilities to perform image recognition based on deep neural network inference at the edge. ABI Research covers some of the key vendors in this market in the Industrial and Commercial Robotics Market(PT-1843) report. This is a space that has seen massive investment in the past few years, with the emergence of Chinese chipset vendors, such as Cambricon and Deep Vision. It will be interesting to see if these chipsets find their way into the AMRs of Chinese robotics companies.

As compared to machine vision, it is still early days for machine learning in e-commerce fulfillment robots. A robot with a mature cognitive system will require localized training and inference of machine learning algorithms. At this stage, however, vision-based machine vision on e-commerce robot is still based on inference using onboard resources, with training being performed in the cloud.

Last but not least is last-mile delivery. Almost all major e-retailers and third-party logistics service providers have demonstrated some versions of automatic last-mile delivery using AMRs or drones. If done right, claims that this would reduce the cost of delivery by 70%. However, the technology remains largely at the testing and experimental phase. For progress to be made, a more robust localization and navigation technology must be developed, similar to those developed by Brain and Seegrid, allowing autonomous platforms to move in highly unstructured environments.