Wojciech Matusik, MIT professor and co-founder of Inkbit, describes a new computational method of packing a container in the most efficient way possible. 

The methodology is built around an algorithm that claims to solve the centuries-old problem of how best to fit multiple objects into a container. The “bean-packing” problem can be found in many industries, Matusik says, but old methods were “extremely slow” and could take hours to arrange hundreds of items optimally.

The new method developed at MIT employs “spectral packing,” which first works out an order of solid 3D objects to fill a container, then breaks a visual representation of both the container and its contents into “voxels,” or tiny cubes within a 3D grid. To figure out available space within the container, the algorithm computes a “collision metric” for each voxel, which identifies any place where one object overlaps, or “collides,” with another. An object can only be placed in locations where the overlap is zero.

Matusik says the methodology was first developed to optimize the placement of parts within a tray for additive manufacturing, or 3D printing. “We wanted to design an algorithm and system that would allow you to make machines as productive as possible.”

But the algorithm can also be applied to much larger environments, such as the packing of hundreds of objects into a shipping container. It could optimize the movement of freight across multiple modes, both domestic and international — “even things that go into orbit,” Matusik says.

Of course, a visual representation of what amounts to a 3D jigsaw puzzle needs to be adjusted to reflect the real-world nature of container packing, accounting for such things as the size, weight, and nature of the objects to be shipped, as well as the desired order of delivery. “These are relatively simple additional objectives that can be laid on top of the existing algorithm,” Matusik says.