Automation at present still limited
Anderson claims that many have the incorrect perception that there will soon be fully automated warehouses. However, Anderson notes that even the task of picking out a single product from a bin without damaging other products or picking up multiple products in error is still years away. Anderson said that Amazon was exploring a number of different technologies to automate the numerous steps needed before a package was sent to shoppers.
While the future of Amazon’s logistics network will no doubt involve AI and robotics it is still an open question when AI-powered machines will even do the majority of work. As it is now, robots are good at doing quite specific repeatable tasks that they are precisely programmed to do. In order to get the robot to do anything else requires expensive time-consuming reprogramming. Robots that do many different tasks and are able to see and understand their environment are very much at the research and experimental stage. Such hardware does not yet exist in the commercial realm.
Human eyes and hands superior to present robots in a warehouse
A robot can help make a microchip and the body of a Tesla EV but it still is not capable of doing the human tasks required in an Amazon warehouse. It is difficult to train robots to see the world and grip items in the way that human hands and eyes can. However, robots are now beginning to gain levels of vision that will soon approach that of human sophistication.
The “picking challenge”
Amazon encourages robotic development by having an annual picking challenge. Robots compete to see which is best at picking one object and moving it successfully to another part of the logistic chain without damaging the items or picking up multiple items.
Other companies pursuing AI robotic research
Other companies than Amazon are also making progress in the field. UC Berkley’s robotics lab has a new low-cost robot with a pair of humanoid type arms that can perform complex tasks such as folding a towel and has a good AI-powered vision system up to the task.
OpenAI a research lab has been using an AI training technique called reinforcement learning to teach a robotic hand more precise movements of the type needed to replicate a human hand in warehouse work.
Finally Kindred of San Francisco is making an arm called KindredSort that has been deployed in retailer Gap’s warehouse. However, the hand combines automation with human direction but with that the hand can do product picking.
Amazon’s many warehouses make limited use of robots
Amazon is reported to have 110 warehouses, 45 sorting centers, and about 50 delivery stations just in the US alone. They employ about 125,000 full-time warehouse workers. Only a small fraction of the work is done by robots. At present, robots are too imprecise and clumsy unless subject to a great deal of training. While research is narrowing the gap with human workers, robots that could challenge human skills are not yet commercially available.
Amazon does uses small Roomba-shaped robots just called drives. Roomba is a robotic vacuum cleaner. The drives deliver large stacks of objects to human workers and are programmed to follow set paths around the warehouses. The drives are shown on the appended video.