Behavior-Aware Multi-Modal Trip Planner: Combining GTFS, OSM, and Reinforcement Learning for User-Centric Routing

Abstract

Urban public transportation systems are widely used, yet most route planning tools primarily optimize objective factors such as travel time or distance, often overlooking subjective aspects of the travel experience such as comfort, crowding, and walking tolerance. As a result, routes that are optimal by technical criteria may feel stressful or impractical to users, limiting the effectiveness of transit systems. This paper proposes a behavior trait-aware multimodal trip planning framework that takes into consideration the personal characteristics of each user. In this regard, the proposed system integrates scheduled transit data, street-level network data, and ride-hailing connectivity within a unified network. In addition, user traits such as urgency, crowding, and willingness to walk are considered within the routing process, affecting both perceived travel cost and routing strategy. The problem of routing is modeled as a reinforcement learning problem using a policy-based optimization approach (Proximal Policy Optimization, PPO), in which the agent is aware of its location, the behavioral characteristics of the user, and the available travel options, in addition to receiving feedback based on travel efficiency and preference alignment.