Online videos may give you an impression of full automation at Amazon warehouses, but you are seeing only the partial picture. What the broad application of robots does not show you is Amazon's unspeakable stress.

Why fully automated warehouses at Amazon are still a decade away?

A Reuters report has revealed that Amazon employs over 125,000 full-time staff on the floor of its U.S. facilities, which include 110 warehouses, 45 distribution centers and around 50 delivery centers, while only a minority of the work is performed by robotics. Amazon is also constantly troubled by reports on its abuse of workers' rights.

Warehouse workers complain that they are treated like robots - the distance one needs to move around the warehouse in one shift feels like a marathon, not having time to go to the bathroom, not to mention the fact that they are being constantly monitored by an automated employee tracking system at any given time.

Wokers operating in an Amazon warehouse (Image credit: Leon Neal/ VCG via Getty Images)

Poor working conditions have led to unusually high turnover rates. Amazon warehouses have to recruit 100,000 new workers each season to balance its workforce.

In fact, and despite being criticized as ruthless, Amazon CEO Jeff Bezos is far more eager to see the coming of full automation than are his warehouse employees, whose salaries ensure the company's fulfillment centers remain Amazon's most expensive facilities.

Yet Amazon is unable to operate without its 100,000 disgruntled, non-robot workers.

The US$775 million acquisition of KIVA System in 2012 introduced into Amazon warehouses orange robots capable of carrying the mobile shelves to the pickers themselves, representing the most unique view inside the facilities, and boosting order placement efficiency.

However, robots are still unable to handle the entire process. Human laborers are still required to manually complete tasks such as removing products from shelves or sorting and packing orders.

According to Scott Anderson, director of Amazon Robotics Fulfillment, even the simple process of identifying an object then picking it up with no previous training on that object requires a series of complex, sophisticated software and hardware that does not yet exist in a commercial fashion.

The action of a robotic arm to pick out a single product or multiple products from a bin as accurately and quickly as a human can without damaging other products necessitates the seamless cooperation of multiple technologies such as machine vision, 3D perception and motion control.

These processes, though seemingly quite simple, entail much more work than a person might imagine. Until recently, Jeff Bezos publicly claimed that only in the next 10 years would there be commercial robots capable of grasping items as reliably as humans can.

In order to accelerate the arrival of full automation at its warehouses, Amazon have been sponsoring the annual Amazon Picking Challenge ever since 2015. It is said that the first event witnessed only 50% of all teams succeed in their attempts to pick and remove products from shelves.

Picking Robots are already 'on duty' in China's warehouses

Facing competitors from all over the world, one MIT Ph.D in Electronic and Information Engineering achieved an outstanding performance during the 'Intelligent Manipulation' competition. As the chief director of technology architecture of the MIT-Princeton team, Yu Guanting consecutively ranked in the top three from 2015 to 2017.

Yu led the MIT-Princeton team in the 2017 Amazon Picking Challenge (Image credit: XYZ Robotics)

In April 2018, as CTO, Yu co-founded XYZ Robotics with Zhou Jiaji (CEO) (Ph.D. in Robotics, School of Computer Science, Carnegie Mellon University) and Xingliang Libo (COO) (Peking University).

In 2016, Zhou Jiaji, as a Ph.D. candidate, was awarded the prize for Best Paper at the world-renowned robotics conference ICRA, in so becoming the third Chinese to win the award as the paper's lead author in the 30-year history of the event.

XYZ Robotics' vision is to "Pick anything. Place anywhere anyway." With hand-eye coordination technology as its core, the company has combined the team's strengths in artificial intelligence, 3D perception and robot manipulation in order to develop visual-picking robot products able to be applied in logistics and industrial scenarios.

Recently, XYZ Robotics secured an US$8 million Series A funding round led by Gaorong Capital and Morningside Venture Capital. Early backers of its Angel rounds, including Sinovation Ventures, New Wheel Capital and Aurora CTO Drew Bagnell (a driverless company in Silicon Valley), also joined the round with further capital injection.

While manual picking and packing remains the greatest efficiency bottleneck inside Amazon's automated warehouses, in the current landscape of logistics automation, the technology of sorting AGVs (automated guided vehicles, or mobile robots), cross-belt sorting machines and AGV forklifts are already well developed. In the first half of 2018, more than 7,000 KIVA warehouse robots were manufactured in China.

However, an appropriate business solution for batch picking at the end of the picking process in warehouses (collecting of multiple orders, summarizing the quantities of each item separately, sorting and distributing the products for customers) is yet to be found.

Currently, humanoid-hand robotic companies are springing up, vying for the 'Holy Grail' in the field of machine intelligence, exploring various implementations that include multi-finger, bionics, software, static electricity and more. SSI SCHAEFER, for instance, is one global logistics automation company which is already on the path to incorporating picking robots into their market offerings.

In December last year, RightHand Robotics, a company that offers robot-picking solutions, secured a US$23 million Series B funding round. The investment in XYZ Robotics, for its part, represents the first large-scale financing in this field in China.

With both entrepreneurs and capital already in the game, why has Mr Bezos made such a gloomy prediction that commercial picking robots are still a decade away?

Part of the reason is that Amazon's warehouses consist largely of mega-warehouses of tens of thousands of square meters, along with hundreds and thousands of SKUs, and where the main storage mode is random hybrid storage, which sets the highest difficulty level for picking robots.

However, for common domestic logistics warehouses holding just thousands of SKUs or of a single category, the conditions for deploying picking robots have already been met.

Amazon has yet to start deploying picking robots, but the technology innovations inspired by the global competition are ready to be applied to smaller, less complex warehouses.

According to Yu, XYZ Robotics has developed two core technologies based on the cumulative experience of the 'Amazon Picking Robot Challenge'.

The first is the machine vision capability to handle tens of thousands of SKUs dynamically updated in the warehouses.

The second is the design capability for flexible fixtures for the robots. For objects of different sizes, shapes and weights, different types of actuators are required at the end of the robotic arm, such as large and small suction caps, and grippers. XYZ Robotics has developed a fast actuator switching system with an execution time of just 0.6 seconds, thus increasing the range of goods the system can handle without sacrificing efficiency.

Picking robots from XYZ Robotics were applied to a warehouse of a Chinese logistics magnate.

In December 2018, XYZ Robotics launched its first automatic batch-picking station able to deal with a variety of categories of mixed goods, which was applied in the Suzhou warehouses of domestic logistics magnates. The robot's picking efficiency proved to be more than 1.5 times of that of a human being, and it did not require any pre-collection of product information.

How soon will a robot be able to "Pick anything. Place anywhere"?

Zhou Jiaji, CEO of XYZ Robotics, admits that the current system to cope with tens of thousands of SKUs is still undergoing rigorous testing. Deployment and testing in complex warehouse environments must be carried out in order to ensure robots can dynamically identify and process in excess of 10,000 SKUs without forgoing accuracy or stability.

On the eve of 'Singles Day' (an annual shopping holiday in China popular with young people) in 2017, a video showing the world's first full-process automated warehouse at JD.com circulated all over the internet. The content of this futuristic video seemed to have captured Bezos' dream, included among it the automated picking robot.

Video of the world's first full-process automated warehouse at JD.com

Was JD.com already ahead of Amazon in warehouse automation? Industry specialists reveal the secret.

Firstly, the JD.com automation warehouses were set in 3C(computer, communication, consumer-electronic) warehouses, where all goods were of the same specification, avoiding the biggest difficulty for picking robots, the diversity of goods.

Secondly, JD.com had reinvented, in accordance with the needs of automated warehouses, the full-process system from dispatch to three-dimensional warehouse manufacturer design to product specification, which was so costly that it was almost impossible to be promoted to other companies.

XYZ Robotics, on the other hand, takes into account the current situation of small and medium-sized domestic warehouses. On account of the cost structure - high rent, high trunk transportation cost and low labor cost - the scale of domestic warehouses is smaller and more intensive. Accordingly, they are still dominated by manual operations, and AGVs have not been widely deployed.

Therefore, XYZ Robotic's current picking robots areplacedin manned warehouses.

Zhou Jiaji says there are several types of workstations in the warehouses - handling, packing, picking and sorting. XYZ is responsible for sorting, its robots operating in cooperation with workers from the other processes. "There is a picker who picks goods from the shelves, then our robots will do the secondary sorting.

XYZ Robotics' intelligent solutions

Compared to the full-process reengineering of JD.com's automated warehouses, XYZ Robotics does not need to disrupt the original workflow of the warehouses. On account of possible inconsistencies resulting from the rhythms of upstream and downstream workers, XYZ Robotics will add some cache, so the robot can still work normally in the case of variations in working efficiency.

XYZ Robotics' customers are primarily warehouses of leading Chinese brands, e-commerce firms and logistics companies. According to Zhou, "A number of leading brands have stronger payment capacity than e-commerce companies, and many customers have a strong willingness for automation."

"The cost structure determines the possibility, forms and proportion of automation." Labor costs are still comparatively low in China, with many warehouses anticipating automation equipment to pay for itself within just two years or even 18 months.

This is also the current goal of XYZ Robotics. The cost of picking robots is expected to be further reduced through the standardization, manufacturing and localization of industrial robots, electrical components, sensors, conveyor belts and other components.

Meanwhile, in order to improve, as soon as possible, the ability of picking robots to handle expanded scales of SKUs, XYZ Robotics has constructed its own small unmanned warehouse. The SKU is updated non-stop, 24/7 to test both the working performance of the picking robots, as well as find the bugs that may occur once the scale of SKUs grow.

Of course, "Pick anything. Place anywhere anyway" is not just a matter of logistics - according to Zhou, in all subsectors of the industrial sector, there are many people doing similar things, all with hand-eye coordination technology at their core. XYZ Robotics is more inclined to output software as a service to integrators - in doing so, it can empower the industry in semi-structured automation scenarios with the capability of hand-eye coordination.

-----------------------------------------------

This article was edited by @Dido Pang . The original Chinese version was written by Zhang Yuan. 

Follow us on Twitter @tmtpostenglish, Medium @TMTPOST and Facebook @TMTPOST.