Let's put more intelligence in material handling systems

Scroll through any news feed you choose, and you will find articles that praise the power and promise of artificial intelligence and machine learning algorithms to transform our society. Artificial intelligence is both the most underrated and overrated technology; overrated to the extent that people attribute super intelligent properties to it and underrated in the sense that we certainly cannot fully appreciate how it will work its way into many facets of our lives. AI has already completely transformed certain specific problems. Check your Amazon recommended purchases or notice the ads Facebook chooses to serve you if you have any doubts. AI is also well on the way to delivering autonomous driving capabilities for all sorts of vehicles and material handling equipment. There is a lot to be excited about.

For all the progress in other areas, where is AI being applied to complex controls decisions in modern distribution centers? In a typical DC filled with conveyor, merges, and sorters, the controls decisions are being made instantaneously based on input from sensors and barcode scanners. Those decisions to turn a conveyor on or off are highly pre-determined; every scenario that can happen is represented in the PLC code. 

There is very little to be gained from trying to apply extra intelligence to these decisions. However, there are an increasing number of systems where extra intelligence can significantly improve performance. These are systems where multiple pieces of equipment can be assigned to any number of potential tasks. Imagine a building with thousands of orders to be fulfilled and hundreds of robots available to bring items to many different stations for picking and shipping. In that not-so- hypothetical scenario, there are very good and very bad ways to assign robots to items. Extra intelligence applied in this context can be the difference between fantastic return on investment and unacceptable returns.

"Big data in   business has   been fuelled in the past five years thanks to digitalisation, process   automation and   the rapid   adoption of the industrial Internet of Things (IIoT)"

Yet often the controls algorithms applied in these situations look more like simple conveyor controls than intelligent algorithms. Developers rely on heuristics and rules of thumb. A typical rule might be to assign the closest idle robot to the item. That sounds logical enough. But will that simple rule ensure that we avoid congestion, keep all of the pack stations utilized, or minimize the total distance travelled by the robot fleet? Certainly not. Yet so many multi-agent, multi-task systems deployed in the world today rely on exactly these types of simplistic heuristics. It is well known that humans are great at pattern recognition and instinctively make good decisions on some types of problems. It is also well understood that humans have cognitive biases that cause us to get it wrong at times. Asking a computer to follow rules-of-thumb-based heuristics is not much different than asking a super-fast human to make all of the decisions. Contrast that with a deep learning algorithm that can adjust its strategy slightly with every movement of a tote and every completion of an order.

The autonomous mobile robot vendors are farther along this curve. My challenge is to the vendors of systems with multiple ASRS cranes, shuttles, pick stations, and carousels where there are many potential execution paths at any given time. Apply AI to your systems and see how much better your performance can be.