Warehouse of the Future: The Power of Compound Learning and Robotics

By Dwight Klappich | July 02, 2019 | 0 Comments

Supply ChainPower of the Profession

In the 1964 film, “Mary Poppins,” Michael is encouraged by his father, Mr. Banks, to “Invest your tuppence wisely in the bank/Safe and sound/Soon that tuppence safely invested in the bank/Will compound.”

Michael instead gave his money away to feed the birds. Notwithstanding this lapse in judgment, compound interest is a great thing since money grows faster over time because it continually earns interest on top of interest.

Albert Einstein has been quoted as saying that compound interest is the “eighth wonder of the world.” “He who understands it, earns it; he who doesn’t, pays it.” Needless to say, both Mr. Banks and Einstein recognized the power of compounding.

So if compound interest is good and powerful, what about knowledge? Can the principles of compound growth have the same geometric impact on knowledge that it has on money? Can learning follow a similar geometric growth trajectory? Can the new directly benefit from the knowledge of the old?

Figure 1 shows a common approach to measuring the impact of change on an individual or organization. Change is not a linear progression. It starts with breaking away from the current state (status quo), which causes uncertainty and turmoil (disruption) that eventually leads to looking for solutions (exploration) and finally moving forward with the new stage (rebuilding).

In Figure 1, the gray lines represent the normal change curve for, say, a new hire who needs to learn and master a new job (Individual A). Performance declines initially as the individual has to learn and adapt to something new. Over time, things will normalize and improve as knowledge and experience advance. However, each time there is something new, the organization starts from scratch and goes through this process all over again such as bringing on new hires Individuals B and C.

190701 July Graphic 1 Web

Compound learning is the knowledge equivalent of compound interest. If compound interest can be viewed as “interest-on-interest,” then compound learning can be viewed as “knowledge-on-knowledge,” where both can make an amount grow at a faster rate. Figure 2 shows how knowledge can grow geometrically over time by building on top of everything learned before.

In warehousing, the rapid emergence of robotics and embedded supporting technologies like artificial intelligence (AI) and machine learning (ML) are poised to change this calculus dramatically. Phrases like “train one, train them all” are no longer far-fetched, they are reality.

The majority of autonomous mobile robot (AMR) companies are able to train one robot and then share this knowledge across all of their similar robots. For example, with AMR vendor Seegrid, users can put the AMR in “training” mode via an onboard interface. In this mode, as the user drives or walks the AMR through tasks, the AMR records its assignments. The learned environment can then be shared with the entire fleet. When adding a new AMR, it starts as knowledgeable as every other unit already in place. For humans it might take weeks for a new employee to learn their way around the warehouse, yet the AMR is up to speed in minutes.

190701 July Graphic 2 Web

While learned autonomous navigation is powerful, compound learning applied to more complex tasks will be even more so. While nascent, multiple robot companies like RightHand Robotics, Kindred and Berkshire Grey have sophisticated embedded AI that is self-learning. Each of these vendors offers some form of robotic picking technology that combines multiple software and hardware technologies. These include robotic arms, grippers, vision systems, and AI software that allow the robot to autonomously identify, pick up and move a variety of items in multiple positions and orientations.

Compound learning comes as the AI software learns from each interaction and the algorithms improve over time. The vision systems get better at recognizing items quicker and more accurately. The arms and grippers get better at adapting to items in unusual positions and developing ways to pick these up. As knowledge grows, the arms can move faster and be more accurate, speeding up the entire process.

Compound learning is manifesting in many ways. First, within an organization all the robots from one vendor can share knowledge and grow collectively. Second, knowledge within a vendor can grow as it learns from all of its customers, with this knowledge shared across its entire customer base. Finally, since many of the vendors use common technologies (e.g., cameras, lasers, sensors), knowledge can be shared across the industry.

Much like the famous Spider-Man quote, “with great power comes great responsibility,” the same is true with compound learning. Compounding the effects of good behaviors will be powerful and beneficial, while compounding the effects of bad behaviors will likewise be powerful, yet devastating.

Gartner is exploring compound learning and its impact through the concept of machine culture — how to manage large numbers of intelligent machines. Seeing how learning sometimes can be very unpredictable, we find that it is difficult to predict how large numbers of machines from different origins and designs (machine cultures) will interact, especially in an emergency.

Warehouse organizations considering robotics should recognize the power of compound learning, but remain cautious in conceding too much control to any black box. Work with your vendors to understand how their solutions learn and evolve, and then institute processes to monitor changes as they occur. Make sure the data you feed the systems is the data you want these systems to learn from.

Dwight Klappich, VP Analyst, Supply Chain Technology, Gartner

The Gartner Blog Network provides an opportunity for Gartner analysts to test ideas and move research forward. Because the content posted by Gartner analysts on this site does not undergo our standard editorial review, all comments or opinions expressed hereunder are those of the individual contributors and do not represent the views of Gartner, Inc. or its management.

Comments are closed