We solve your palletization problems
Getting a robot to stack boxes on a pallet hardly seems challenging - at least, if they are all the same size. However, automating stacking becomes much more complex when presented with a random row of items of different sizes. In that case, the automated stacking system must be able to make a reasoned decision on the best location to place an item on a pallet...
"Pallet Loading Problem": Categories
In pallet automation, we distinguish between two categories:
- Manufacturers Pallet Loading Problem (MPLP): Manufacturing companies typically go through a large quantity of one particular product. This means that the dimensions and weights of products are known in advance, enabling standardised and optimised palletisation.
- Distributors Pallet Loading Problem (DPLP): Distributors such as parcel services face high variability in products and need to respond to the dynamics of daily deliveries. In this case, a robot cannot be programmed once, but the locations on the pallet must be adjusted based on the boxes offered.
Aside from these 2 categories, it is also important whether the sizes of the products are known in advance:
- Offline Positioning: In this case, all products that should end up on the pallet are known in advance. With data known in advance, the pallet layout can be optimised to ensure maximum efficiency and stability. Note that this can require a lot of processing power.
- Online Positioning: A typical example is the conveyor belt on which products to be loaded onto a pallet are placed. 1 or a limited number of products are offered each time. Decisions must therefore be made based on the knowledge at that moment, and not on the basis of all the products that will have to go on the pallet. A product must be placed ‘as best as possible’.
Decision algorithm for flexible palletising
Flanders Make developed a new decision algorithm to let robots stack random boxes in a stable way. It can automatically calculate the optimal way to stack boxes of different shapes and heights, without knowing the order or dimensions.
To determine the optimal position of a box on the pallet, we looked at how a person would stack these boxes on a pallet. For example, he is more likely to place large boxes on the corners, and medium-sized boxes on the edge of the pallet. The smaller boxes will then be in the middle. In addition, he will place boxes against each other as much as possible, trying to avoid too big gaps where no box can be placed, or sometimes leave a space free because a small box can still be placed between them.
Based on these rules of thumb, we developed an algorithm that uses heuristics to assign scores to all possible positions. The position with the highest score is the one chosen. The following steps are performed:
This algorithm is built like a pipeline, so steps can easily be added or omitted, or replaced with another version.
The different steps have parameters that can be adjusted so that the algorithm can be fine-tuned. For example, a lower score can be given to placing a large box on a corner. That way, the final result can be tuned to the needs of a specific environment.
A further optimisation can be done if multiple boxes are detected at the input. In that case, scores are calculated for all detected boxes. Based on these scores, and the layer on which the boxes will be put, the best box is taken. An important consequence of this is that smaller boxes are better used to fill spaces.
Concept: Online palletisation of a limited number of variants
Our research focused on online palletising of a limited number of variants. It came about as part of the Coock project CEROS.
We used the following variants in the ground planes of the boxes:
- Small: base = 50mm x 75 mm
- Medium: base = 75mm x 100mm
- Large: base = 100mm x 150mm
As the ground surfaces are multiples of each other (Small fits twice in Medium), this allows combinations that stack well. By choosing the boxes in this way, it does allow enough combinations of boxes to arrive at a properly stacked pallet. The bases chosen are similar to A4, A5 and A6.
The palletisation algorithm we developed does not by itself take into account the fact that boxes are multiples of each other, so this is not a necessity. Besides variations in ground planes, there are 3 different heights namely 40mm, 80mm and 120mm.