Physical question, virtual answer: Optimized real-time physical simulations and physics-informed learning approaches for cargo loading stability

Cargo stability is a crucial requirement for safe cargo loading and transport. Current state-of-the-art approaches simplify cargo loading to an idealized static problem and employ geometric- and force-based approaches. In this research, we model cargo loading stability as a dynamic problem and propose two approaches. We use (a) a physical simulation using a real-time physics engine fitted for cargo loading and (b) a physics-informed learning model trained on cargo loading data. Both approaches are capable of handling dynamic physical behavior, either explicitly through simulation, or implicitly through training a recurrent neural network on physically-biased sequential cargo loading data. Given our two objectives of maximal accuracy and minimal runtime, our benchmarking results show that our approaches can outperform current state-of-the-art static stability methods in terms of accuracy depending on the complexity scenario, but consume more runtime.