Existing robotic technology in the warehouse has provided beneﬁts in horizontal transportation. Instead of the worker going into the racking or storage area to get stock, the robotic device does the retrieval. It turns traditional warehouse inventory transfer and fulﬁllment on its head.
Automating in two dimensions
In effect, its main efﬁciency gains have been to automate two of three warehouse dimensions – horizontal transportation. Horizontal transportation technology has been used for a long time in warehouse environments. It began with ﬁxed-path AGVs (automated guided vehicles), which then were upgraded to SGVs (self-guided vehicles).
The most recent innovations employ improved AGV technology that provides mobility without requiring changes to existing warehouse infrastructure. This enhanced technology is combined with a command and control system that permits the AGVs to dynamically reroute themselves. When used with advanced sensory and guidance systems, they can be deployed for warehouse inventory transfer operations.
Like historical AGVs, this ﬁrst generation inventory transfer transports what I will call “pallet sized” mobile racking or shelving units to ﬁxed picking and packing workstations.
At the work stations human beings take over to pick and pack the individual items from the shelves. Once this is done the shelves are returned to their original location by an AGV.
The problem with this type of operation is that it still has inefﬁciencies. The inventory transfer operation carries surplus stock. Because they are, in effect, moving the whole racking unit, the robots carry surplus quantities of the item to be picked and carry different items (SKUs) on the other mobile shelf slots that are not required for the speciﬁc pick
Adding a new dimension
The mobile robots are transporting batch-sized quantities, when both smaller quantities and more precise picking of items that have been ordered are actually required. To achieve this our next generation of robots must work in the third dimension – height. For the original AGV warehouse robot, the principal design criterion was to select any shelf and deliver it to any pick-pack station at any time.
In the next generation of robotics warehouses will want robots to pick cases, or even more granular, broken-case items individually, as compared to pulling a full bay of shelving.
In effect, the demand will be for robotic systems that provide mobility and combine it with manipulation and vision. The challenge is to use computer vision and machine deep-learning technologies to pick items one at a time and then package them.
Command and control software
With all the focus on robotics hardware and robots it is easy to forget that underlying all the visible technology is an extensive array of command and control software.
- navigation software;
- collision avoidance software for both robots and humans;
- warehouse robotic control software;
- software to recharge the robotic batteries;
- integration software to integrate the robots into inventory control, customer ordering and fulﬁllment systems, and inventory replenishment software.
Another robotic design challenge is how to handle the many different types of objects that must be picked. The complicated way to solve this challenge would have been to write extensive rules teaching a robot how best to grasp a speciﬁc object or shape.
But, given the number and variety of possible shapes a picking robot might encounter on the job, it made more sense to enable the robot to learn. It can thus study a large number of 3D images and learn to recognize which types of grip would work for different shapes. This allowed it to ﬁgure out suitable grips for new objects. Recently, robotics researchers have increasingly used a powerful machine-learning approach known as deep learning to improve these capabilities.
One of the more interesting questions is what will be the most productive robotic process. To date it has been bringing the stock to the human picker/packer.
In the future, with improved robotic capabilities, enhanced articiﬁal intelligence, machine vision and learning, the question is whether will we return to the picker-to-goods model, except this time the picker will be a robot who collects items for each individual order, and probably packs them too.