Our early research on high-density storage system suggested a natural tradeoff between storage density and throughput—the higher the density, the lower the throughput. Is there a way to get both?
The GridStore system combines puzzle-based movement and decentralized control to produce a high-density storage system that conforms to the plug-and-play paradigm. The system is built on the same hardware model as the Flexconveyor—a unit-sized conveyor segment that conveys in the four cardinal directions, has communications ports on all four sides, and has its own controller. Each conveyor is capable of talking to its four neighbors, and no others.
Control of the GridStore system is completely decentralized. At each time step, conveyors assess their current state (Unoccupied? Occupied? Occupied with moving item?), then enter a round of electronic negotiations with their neighbors to determine which direction each conveyor will convey. The result is the global behavior you see here:
Black items in the video have been requested by an external source, such as an order picking workstation or a shipping schedule. Blue items are returning to replenish the system. Each blue item has a target row, to which it travels and then awaits a request. Target rows match the home rows of departing items, thereby maintaining system balance.
Throughput in the GridStore system can be controlled by the number of rows and columns in the grid, by the number of empty cells k, and by the number of requested items (WIP), which we keep constant in a sort of “constant work-in-process” (CONWIP) system. Throughput goes up as the number of empty cells increases, and it also goes up as WIP increases. In ordinary storage systems subject to congestion, the marginal increase in throughput goes down as WIP increases; the opposite is true in GridStore, for low to medium levels of WIP.
As the number of active requests (WIP) increase, throughput increases non-linearly because (1) additional WIP tends to empty out the lower rows, clearing them for requested items and (2) requested items are more likely to form “slugs” which move together very efficiently. For very high levels of WIP (60+ in the plot), most items are requested and the system becomes “river-like.”
If you would like to learn more about the algorithms behind GridStore, please email me with the form below.