Naturalistic Routing Using Inverse Reinforcement Learning

This disclosure describes techniques, referred to as naturalistic routing (NR), that improve the quality of routes found by map applications by learning from users’ real-world navigation actions, accessed with user permission. The techniques leverage the principle that users, in the aggregate, tend to travel on optimal routes to reach their destinations. A machine learning model is trained using inverse reinforcement learning and provides routes that are optimal by the users’ definition of optimality, as determined from a dataset of navigation actions