DataSP: A Differential All-to-All Shortest Path Algorithm for Learning Costs and Predicting Paths with Context
This addresses path planning challenges in domains like robotics or navigation by enabling efficient learning from large trajectory datasets, though it is incremental as it builds on differentiable combinatorial solvers.
The paper tackles the problem of learning latent transition costs on graphs from trajectory demonstrations with contextual features, introducing DataSP, a differentiable all-to-all shortest path algorithm that outperforms state-of-the-art methods in predicting paths on graphs.
Learning latent costs of transitions on graphs from trajectories demonstrations under various contextual features is challenging but useful for path planning. Yet, existing methods either oversimplify cost assumptions or scale poorly with the number of observed trajectories. This paper introduces DataSP, a differentiable all-to-all shortest path algorithm to facilitate learning latent costs from trajectories. It allows to learn from a large number of trajectories in each learning step without additional computation. Complex latent cost functions from contextual features can be represented in the algorithm through a neural network approximation. We further propose a method to sample paths from DataSP in order to reconstruct/mimic observed paths' distributions. We prove that the inferred distribution follows the maximum entropy principle. We show that DataSP outperforms state-of-the-art differentiable combinatorial solver and classical machine learning approaches in predicting paths on graphs.