Massively Scalable Inverse Reinforcement Learning in Google Maps

TL;DR

This paper introduces scalable inverse reinforcement learning techniques for planetary-scale route optimization in Google Maps, achieving significant improvements in route quality and addressing computational challenges at unprecedented scale.

Contribution

It presents novel scaling methods, a new IRL algorithm (RHIP), and demonstrates the largest real-world IRL study with substantial performance gains.

Findings

16-24% improvement in route quality

Achieved planetary-scale IRL with hundreds of millions of states

Identified trade-offs between deterministic and stochastic planning

Abstract

Inverse reinforcement learning (IRL) offers a powerful and general framework for learning humans' latent preferences in route recommendation, yet no approach has successfully addressed planetary-scale problems with hundreds of millions of states and demonstration trajectories. In this paper, we introduce scaling techniques based on graph compression, spatial parallelization, and improved initialization conditions inspired by a connection to eigenvector algorithms. We revisit classic IRL methods in the routing context, and make the key observation that there exists a trade-off between the use of cheap, deterministic planners and expensive yet robust stochastic policies. This insight is leveraged in Receding Horizon Inverse Planning (RHIP), a new generalization of classic IRL algorithms that provides fine-grained control over performance trade-offs via its planning horizon. Our contributions culminate in a policy that achieves a 16-24% improvement in route quality at a global scale, and to the best of our knowledge, represents the largest published study of IRL algorithms in a real-world setting to date. We conclude by conducting an ablation study of key components, presenting negative results from alternative eigenvalue solvers, and identifying opportunities to further improve scalability via IRL-specific batching strategies.