If the last 10 years have been revolutionary for the supply chain industry, the next decade will be even more dynamic. According to research by Reportlinker, the global market of AI solutions for logistics will be growing by a CAGR of 39.6% between 2020 and 2027. Such growth in investments will mean constant changes in warehouse operations and processes, progressively going towards complete automation. At the same time, progressive upgrading of warehouses will be essential for many companies to avoid waiting for a next-generation warehouse to get any further automation. With this so-called “brownfield” approach, the evolution of operations will be easier to control, yet still quite challenging.
To prepare for this, let us look at some good practices that can help make warehouse automation more friendly. I will take real-life examples from robotic picking, which is the next big step in intralogistics, will surely make a good lesson.
This may seem obvious but the first step for increasing automation is to analyze which part of the process, or which subset of SKUs should be the focus of additional automation. Robotic solutions are improving every day, but they cannot yet handle all kinds of use cases or all types of items. It would be naïve to expect that a new, cutting-edge technology can be adopted and improve the flow of all your SKUs overnight. Focus is essential to create an effective environment for iterating and improving the operations progressively. This goes for any kind of innovation that is being implemented in the warehouse.
For example, in robotic piece picking, for each new customer it is essential to analyze the processed SKUs to determine which ones will be pickable by the robot and which ones will not. This is the first phase of all our projects at Nomagic, and it is done with a dedicated tool which we call the ‘Product Scout’. Product Scout essentially takes pictures of SKUs sold in a given period and analyzes their shape, position and characteristics to determine whether they are suitable for robotic picking.
Focusing on a smaller number of SKUs at once brings the advantage that we can quickly achieve target KPIs such as throughput and autonomy and make the solution viable for sustained operations. Then, after this initial calibration and once the first robot is set in the right direction, we can move on to further categories of SKUs and progressively deploy more robots.
There is a lot of innovation going on in the intralogistics industry and surely you will often be amazed at what you see at various conferences and events. But – looking again at our example – demos of robotic picking at conferences have no value, especially if they come from vendors who cannot show real production experience. An isolated lab environment will not tell you much about the actual performance of the product in a warehouse. This is true because each use case is different, and exception cases vary from one warehouse to another. Where you pick from, what kind of items are handled or how do you place or pack items – the number of combinations is infinite.
Because of that it is key to first identify your needs very well and then look for a vendor who has already got some production experience with at least comparable use cases. Otherwise, if you both start from scratch, a lot of time and effort will be wasted before you get some really good results.
However, one problem with testing in production early on is the cost of such tests. You want to identify the right vendor and together define processes to be able to implement a test without having to invest hundreds of thousands of euros. Exactly this pushed us at Nomagic to develop a dedicated system to have our robots interact directly with the operator screen at a goods-to-person station, hence removing the need for WMS integration during the pilot phase. This way, we can use this stage to focus mostly on ramping-up the system on the selected set of SKUs (back to Rule #1!) and work on the WMS integration later on.
Finally, the most important feature of a completely automated system is its autonomy. But as we integrate more and more systems that use machine learning and tend to improve over time, autonomy should not be taken for granted and is rather an outcome of the integration and adaptation process. As explained with the process for selection of SKUs (see Rule #1), the picking system can only become autonomous gradually. And the main KPI used to measure this autonomy is the intervention rate or how often a person has to attend to the system.
Given the gradual improvement of the system, the initial setup and ramp-up may take some time and interventions from you or your vendor. The remedy is not to find a supplier who can promise that this will not happen and its machine learning system will be correct from day one. The best that can be done is establishing remote operations that will remove the need for onsite interventions. This means investing in a process by which people can intervene on the automated system (for us a robot) without having to go onsite next to it.
The way we are optimizing it at Nomagic is by having our own remote operators, called watchers, remotely monitor the movements of any robot at a client site. Our proprietary robot-watcher communication system alerts the watcher whenever the robot encounters a problem and enables them to control the robotic arm and remove the obstacle or place the object in a different way. Each time multiple photos are being taken and detailed positioning data is being saved for engineers to analyze after the fact and eventually make the system autonomously resilient to the given case. A vast majority of issues can be handled this simple way, which reduces drastically the number of onsite interventions. This in turn results in much better uptime and progressively makes the system 100% automated.
The end goal of warehouse automation, which is the lights-out warehouse, is not something that will happen tomorrow. However, when it does, it will not be in a lab of a robotics startup or in the development center of an integrator. It will have to happen in a production warehouse with real products, real project engineers, for a selected part of the process (Rule #1), after careful validation onsite (Rule #2) and thanks to support from remote operations (Rule #3). The iterative nature of such solutions means that those who give more time to adapt a solution to their needs will reap the most benefits. I hope that the approach that I outlined here will help to make this process relatively smooth and can serve as a good benchmark for introducing end-to-end warehouse automation especially for existing warehouses that can still be further automated.
This post is also available in : Français (French)