Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TRIP-PAL: Travel Planning with Guarantees by Combining Large Language Models and Automated Planners

Tomas de la Rosa, Sriram Gopalakrishnan, Alberto Pozanco, Zhen Zeng, Daniel Borrajo

TL;DR

TRIP-PAL addresses the gap between LLM generated travel plans and hard feasibility constraints by marrying LLM-driven information extraction with a formal automated planner. The method translates user goals and POI data into planning problems and solves them to guarantee validity while optimizing a utility function. Experimental results across diverse city scenarios show TRIP-PAL outperforms GPT-4 in plan quality and constraint satisfaction, especially as the number of POIs grows. The work demonstrates the value of hybrid LLM-planner pipelines for travel planning and establishes a benchmark for oversubscription planning in this domain.

Abstract

Travel planning is a complex task that involves generating a sequence of actions related to visiting places subject to constraints and maximizing some user satisfaction criteria. Traditional approaches rely on problem formulation in a given formal language, extracting relevant travel information from web sources, and use an adequate problem solver to generate a valid solution. As an alternative, recent Large Language Model (LLM) based approaches directly output plans from user requests using language. Although LLMs possess extensive travel domain knowledge and provide high-level information like points of interest and potential routes, current state-of-the-art models often generate plans that lack coherence, fail to satisfy constraints fully, and do not guarantee the generation of high-quality solutions. We propose TRIP-PAL, a hybrid method that combines the strengths of LLMs and automated planners, where (i) LLMs get and translate travel information and user information into data structures that can be fed into planners; and (ii) automated planners generate travel plans that guarantee constraint satisfaction and optimize for users' utility. Our experiments across various travel scenarios show that TRIP-PAL outperforms an LLM when generating travel plans.

TRIP-PAL: Travel Planning with Guarantees by Combining Large Language Models and Automated Planners

TL;DR

TRIP-PAL addresses the gap between LLM generated travel plans and hard feasibility constraints by marrying LLM-driven information extraction with a formal automated planner. The method translates user goals and POI data into planning problems and solves them to guarantee validity while optimizing a utility function. Experimental results across diverse city scenarios show TRIP-PAL outperforms GPT-4 in plan quality and constraint satisfaction, especially as the number of POIs grows. The work demonstrates the value of hybrid LLM-planner pipelines for travel planning and establishes a benchmark for oversubscription planning in this domain.

Abstract

Travel planning is a complex task that involves generating a sequence of actions related to visiting places subject to constraints and maximizing some user satisfaction criteria. Traditional approaches rely on problem formulation in a given formal language, extracting relevant travel information from web sources, and use an adequate problem solver to generate a valid solution. As an alternative, recent Large Language Model (LLM) based approaches directly output plans from user requests using language. Although LLMs possess extensive travel domain knowledge and provide high-level information like points of interest and potential routes, current state-of-the-art models often generate plans that lack coherence, fail to satisfy constraints fully, and do not guarantee the generation of high-quality solutions. We propose TRIP-PAL, a hybrid method that combines the strengths of LLMs and automated planners, where (i) LLMs get and translate travel information and user information into data structures that can be fed into planners; and (ii) automated planners generate travel plans that guarantee constraint satisfaction and optimize for users' utility. Our experiments across various travel scenarios show that TRIP-PAL outperforms an LLM when generating travel plans.
Paper Structure (20 sections, 5 figures)

This paper contains 20 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Travel Planning
  4. Travel Planning with LLMs
  5. LLMs and Planning
  6. Method
  7. Retrieval of Travel Information
  8. LLM-based Tourist Plan
  9. Automated Planning Task
  10. Evaluation
  11. Experimental Setting
  12. Approaches.
  13. Benchmark.
  14. Metrics.
  15. Results
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Components used in travel planning for GPT-4 and $\textsc{TRIP-PAL}$.
  • Figure 2: PDDL definition of the visit action.
  • Figure 3: Utility of the plans returned by $\textsc{TRIP-PAL}$ and GPT-4. Points below the solid diagonal indicate that $\textsc{TRIP-PAL}$ generated a higher quality plan than GPT-4 for the given travel planning task. Points at the horizontal red line indicate that GPT-4 generated an invalid plan.
  • Figure 4: Distribution of the number of POIs visited by each plan, as well as their average, maximum, and minimum utility.
  • Figure 5: GPT-4 ratio of invalid plans (first column), GPT-4 suboptimality ratio (second column) and $\textsc{TRIP-PAL}$ extra seconds compared to GPT-4 (third column) as we increase the number of POIs (first row) and the travel hours $H$ (second row).