TourPlanner: A Competitive Consensus Framework with Constraint-Gated Reinforcement Learning for Travel Planning

TL;DR

TourPlanner introduces a novel framework combining multi-path reasoning and constraint-gated reinforcement learning to enhance travel itinerary planning, effectively balancing constraint satisfaction and user preferences.

Contribution

It presents a comprehensive approach with multi-path reasoning and a gating mechanism, addressing key challenges in travel planning that previous methods struggled with.

Findings

Achieves state-of-the-art performance on travel planning benchmarks.

Significantly improves feasibility and user-preference alignment.

Outperforms existing methods in handling constraints and exploration.

Abstract

Travel planning is a sophisticated decision-making process that requires synthesizing multifaceted information to construct itineraries. However, existing travel planning approaches face several challenges: (1) Pruning candidate points of interest (POIs) while maintaining a high recall rate; (2) A single reasoning path restricts the exploration capability within the feasible solution space for travel planning; (3) Simultaneously optimizing hard constraints and soft constraints remains a significant difficulty. To address these challenges, we propose TourPlanner, a comprehensive framework featuring multi-path reasoning and constraint-gated reinforcement learning. Specifically, we first introduce a Personalized Recall and Spatial Optimization (PReSO) workflow to construct spatially-aware candidate POIs' set. Subsequently, we propose Competitive consensus Chain-of-Thought (CCoT), a multi-path reasoning paradigm that improves the ability of exploring the feasible solution space. To further refine the plan, we integrate a sigmoid-based gating mechanism into the reinforcement learning stage, which dynamically prioritizes soft-constraint satisfaction only after hard constraints are met. Experimental results on travel planning benchmarks demonstrate that TourPlanner achieves state-of-the-art performance, significantly surpassing existing methods in both feasibility and user-preference alignment.