MIT researchers applied deep-learning approaches from vehicle routing to streamline planning trajectories for robots in an e-commerce warehouse. Their method breaks the problem down into smaller chunks and then predicts the best chunks to solve with traditional algorithms.