[Paper Review] Applying Strategic Multiagent Planning to Real-World Travel Sharing Problems

Jan Hrnčíř, Michael Rovatsos|arXiv (Cornell University)|Jan 2, 2013

Multi-Agent Systems and Negotiation16 references3 citations

TL;DR

This paper proposes a three-phase multiagent planning algorithm that uses off-the-shelf single-agent planners and best-response planning to generate individually rational, cost-effective shared travel routes in real-world public transportation networks. Evaluated on UK public transit data, the method scales linearly with scenario size and agent count, achieving up to 53% cost reduction for three-person groups with a 70% timetable feasibility rate.

ABSTRACT

Travel sharing, i.e., the problem of finding parts of routes which can be shared by several travellers with different points of departure and destinations, is a complex multiagent problem that requires taking into account individual agents' preferences to come up with mutually acceptable joint plans. In this paper, we apply state-of-the-art planning techniques to real-world public transportation data to evaluate the feasibility of multiagent planning techniques in this domain. The potential application value of improving travel sharing technology has great application value due to its ability to reduce the environmental impact of travelling while providing benefits to travellers at the same time. We propose a three-phase algorithm that utilises performant single-agent planners to find individual plans in a simplified domain first, then merges them using a best-response planner which ensures resulting solutions are individually rational, and then maps the resulting plan onto the full temporal planning domain to schedule actual journeys. The evaluation of our algorithm on real-world, multi-modal public transportation data for the United Kingdom shows linear scalability both in the scenario size and in the number of agents, where trade-offs have to be made between total cost improvement, the percentage of feasible timetables identified for journeys, and the prolongation of these journeys. Our system constitutes the first implementation of strategic multiagent planning algorithms in large-scale domains and provides insights into the engineering process of translating general domain-independent multiagent planning algorithms to real-world applications.

Motivation & Objective

• To address the strategic complexity of real-world travel sharing, where agents have conflicting preferences and require individually rational joint plans.
• To evaluate the feasibility of applying advanced multiagent planning techniques to large-scale, real-world public transportation data.
• To design and implement a practical, scalable algorithm that ensures stability (Nash equilibrium) and individual rationality in shared journey planning.
• To bridge the gap between theoretical multiagent planning and real-world deployment in transportation systems.

Proposed method

• First, individual travel plans are generated using high-performance, domain-independent single-agent planners on a simplified, abstracted domain that omits detailed timetabling.
• Second, a best-response planner (BRP) merges individual plans into a joint multiagent plan that is a Nash equilibrium, ensuring no agent can improve their outcome by unilaterally deviating.
• Third, the resulting joint plan is mapped back onto the full temporal planning domain to schedule actual journeys using real timetable data.
• The algorithm uses domain-independent planners for initial and merging phases, with a domain-specific timetabling phase to handle real-world connectivity constraints.
• The approach is designed to be extensible, allowing for additional constraints such as mandatory co-travel with specific individuals or time windows at destinations.
• A preprocessing clustering step is suggested for large-scale deployment to group agents by origin, destination, departure time, and preferences, improving scalability.

Experimental results

Research questions

• RQ1 Can state-of-the-art multiagent planning techniques be effectively applied to real-world, large-scale public transportation data for travel sharing?
• RQ2 How does the proposed algorithm scale with increasing numbers of agents and scenario sizes in a realistic domain?
• RQ3 To what extent can the algorithm achieve cost reduction and timetable feasibility while maintaining individual rationality and stability in joint plans?
• RQ4 What trade-offs exist between cost improvement, timetable feasibility, and journey prolongation in shared travel scenarios?

Key findings

• The algorithm exhibits linear scalability in both scenario size and number of agents, with average computation times under one hour for 12 agents in a large UK train network scenario.
• For three-person groups in central UK train scenarios, the method achieves a 70% success rate in finding feasible timetables, with potential cost savings of up to 53% compared to individual travel.
• The approach avoids the exponential blowup in action space typical of centralized multiagent planners by using a decomposition into individual planning and best-response merging.
• Larger groups yield greater cost savings but face diminishing feasibility due to increased difficulty in matching joint plans to actual timetables.
• The timetabling phase is theoretically incomplete due to simplifications, but empirical results show it remains effective in a large fraction of cases.
• The method is adaptable to other domains, with the initial and BR phases being fully domain-independent, while the timetabling phase requires domain-specific adaptation.