Preference Guided Iterated Pareto Referent Optimisation for Accessible Route Planning

Paolo Speziali; Arno De Greef; Mehrdad Asadi; Willem Röpke; Ann Nowé; Diederik M. Roijers

cs.AIApr 1, 2026

Preference Guided Iterated Pareto Referent Optimisation for Accessible Route Planning

Authors: Paolo Speziali, Arno De Greef, Mehrdad Asadi, Willem Röpke, Ann Nowé, Diederik M. Roijers

Organizations: Vrije Universiteit Brussel Brussels, Belgium · Gemeente Amsterdam Amsterdam, Netherlands

Abstract

We propose the Preference Guided Iterated Pareto Referent Optimisation (PG-IPRO) for urban route planning for people with different accessibility requirements and preferences. With this algorithm the user can interact with the system by giving feedback on a route, i.e., the user can say which objective should be further minimized, or conversely can be relaxed. This leads to intuitive user interaction, that is especially effective during early iterations compared to information-gain-based interaction. Furthermore, due to PG-IPRO's iterative nature, the full set of alternative, possibly optimal policies (the Pareto front), is never computed, leading to higher computational efficiency and shorter waiting times for users.

Figures & tables

Figure 1. Interface design for interactive route planning.

Panel of Figure 2 — Figure 2. Iteratively update the objective search space in IPRO (Figure from ( Röpke et al., 2024 ) )

Figure 3. Average maximum utility comparison between PG-IPRO and GPPE for a convex and a concave Pareto front consisting of 30 solutions (noise=0.01)

Figure 4. The seven possibly optimal routes in Osdorp-Midden. The blue marker (above) indicates the origin and the red marker (below) the destination. Each colored path corresponds to one Pareto-undominated solution with objective vector $[d,c]$ , where $d$ denotes route length (in metres) and $c$ the number of crossings. The yellow-green route on the extreme left corresponds to $[928,3]$ , while the orange route extending furthest to the right corresponds to $[1335,2]$ . In the central region, the pink and brown routes in the upper part correspond to $[703,4]$ and $[603,5]$ , respectively, and the red route in the lower central area corresponds to $[586,6]$ . The two routes closest to the lower boundary are the dark blue route $[568,8]$ and the light blue route $[574,7]$ .

Figure 5. Average maximum utility for Osdorp-Midden PF with 7 solutions (noise=0.01)

Travel Planning

See all 21 papers

From Feasibility to Desirability: Plan, Learn, Adapt (PLA) Framework for Personalized On-Device Itinerary Generation
Jul 17, 2026Himel Dev, Tanmoy Sen, Madhusudan Basak +1Travel Planning User Preference Modeling
Tech LLC · University of Virginia · University of Massachusetts Amherst
Customizable and Jointly Optimized Route Planning: A Deep Architecture Enabling Differentiable Shortest-Path Search
Sep 17, 2026Rui Zhao, Chao Chen, Longfei Xu +4Graph-Based Planning Differentiable Optimization
Alibaba Group Beijing, China · Alibaba Group Hangzhou, China
Embark Now: User Demand Oriented Framework for Multi-day Urban Travel Itinerary Planning
Jul 12, 2026Rongbo Qi, Yaqi Zhang, Shijun Yan +3Metaheuristic Optimization
College of Computer Science, Nankai University, Tianjin, 300350, Tianjin, China. · Jiangxi Normal University, Nanchang, 330022, Jiangxi, China. · College of Software, Nankai University, Tianjin, 300350, Tianjin, China.

Jul 17, 2026J/K move · Enter open · S save

From Feasibility to Desirability: Plan, Learn, Adapt (PLA) Framework for Personalized On-Device Itinerary Generation

Himel Dev, Tanmoy Sen, Madhusudan Basak, Bashima Islam

Tech LLC · University of Virginia · University of Massachusetts Amherst

Travel Planning User Preference Modeling

Open paper AskPDF arXiv

Generating personalized trip itineraries is a complex planning task and involves a tension between hard combinatorial feasibility and soft latent desirability. Classical optimization enforces constraints but fails to capture subjective traveler preferences. While learning-based approaches model preferences, they cannot guarantee feasibility. Mobile deployment imposes additional resource constraints on both. To address this, we propose Plan, Learn, Adapt (PLA), a three-stage framework for personalized on-device itinerary generation. The Plan stage builds a heterogeneous ensemble of lightweight planners that produces structurally diverse feasible candidates. From pairwise itinerary comparisons, Learn fits a compact Bradley-Terry reward model that captures emergent schedule properties such as pacing, geographic coherence, and day balance, which per-POI signals miss. Finally, Adapt applies feasibility-preserving local refinement within a device-aware compute budget; every intermediate state is feasible by construction. On 2,519 pairwise human comparisons across more than 100 U.S. cities, the reward-guided ensemble achieves a 67.8% win rate, 11.2 percentage points above the best single planner, with 100% feasibility. Three frontier LLMs, GPT-5, Claude Opus 4.5, and Gemini 3 Pro, achieve 0% feasibility under the same constraints. The reward model generalizes across held-out cities, with a 67.6% mean leave-one-city-out accuracy. In production deployment within FlyEnJoy, PLA increased itinerary completion rates by 91%, with 109.9 ms average on-device latency.

Abstract

Travel Planning

Explore similar work

From Feasibility to Desirability: Plan, Learn, Adapt (PLA) Framework for Personalized On-Device Itinerary Generation

Customizable and Jointly Optimized Route Planning: A Deep Architecture Enabling Differentiable Shortest-Path Search

Embark Now: User Demand Oriented Framework for Multi-day Urban Travel Itinerary Planning