Loading...

Media is loading
 

Description

Proposing air cargo palletizing solutions requires an assessment by a physics engine of whether a solution is physically stable, which can take up a disproportionate amount of computation and, thus, produce a bottleneck in the optimization pipeline. This problem can be tackled by replacing the physics engine with a data-driven model that learns to map proposed packing pattern solutions directly to its stability value. We develop a prototype of such a datadriven model and find that this approach yields practicable results and does so multiple orders of magnitudes faster than a commonly used physics engine.

Share

COinS
 
Jan 17th, 12:00 AM

Predicting Static Stability with Data-Driven Physics in Air Cargo Palletizing

Proposing air cargo palletizing solutions requires an assessment by a physics engine of whether a solution is physically stable, which can take up a disproportionate amount of computation and, thus, produce a bottleneck in the optimization pipeline. This problem can be tackled by replacing the physics engine with a data-driven model that learns to map proposed packing pattern solutions directly to its stability value. We develop a prototype of such a datadriven model and find that this approach yields practicable results and does so multiple orders of magnitudes faster than a commonly used physics engine.