Jiayi Zeng, Yizhe Feng, Mengliang He, Wenhui Lei, Wei Zhang, Zeming Liu, Xiaoming Shi, Aimin Zhou. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Although large language models have enhanced automated travel planning abilities, current systems remain misaligned with real-world scenarios.First, they assume users provide explicit queries, while in reality requirements are often implicit.Second, existing solutions ignore diverse environmental factors and user preferences, limiting the feasibility of plans.Third, systems can only generate plans with basic POI arrangements, failing to provide all-in-one plans with rich details.To mitigate these challenges, we construct a novel dataset RETAIL, which supports decision-making for implicit queries while covering explicit queries, both with and without revision needs.It also enables environmental awareness to ensure plan feasibility under real-world scenarios, while incorporating detailed POI information for allin-one travel plans.Furthermore, we propose a topic-guided multi-agent framework, termed TGMA.Our experiments reveal that even the strongest existing model achieves merely a 1.0% pass rate, indicating real-world travel planning remains extremely challenging.In contrast, TGMA demonstrates substantially improved performance 2.72%, offering promising directions for real-world travel planning.1 sistant for personalized travel planning.

Reward design in reinforcement learning and optimal control is challenging. Preference-based alignment addresses this by enabling agents to learn rewards from ranked trajectory pairs provided by humans. However, existing methods often struggle from poor robustness to unknown false human preferences. In this work, we propose a robust and efficient reward alignment method based on a novel and geometrically interpretable perspective: hypothesis space batched cutting. Our method iteratively refines the reward hypothesis space through “cuts” based on batches of human preferences. Within each batch, human preferences, queried based on disagreement, are grouped using a voting function to determine the appropriate cut, ensuring a bounded human query complexity. To handle unknown erroneous preferences, we introduce a conservative cutting method within each batch, preventing erroneous human preferences from making overly aggressive cuts to the hypothesis space. This guarantees provable robustness against false preferences, while eliminating the need to explicitly identify them. We evaluate our method in a model predictive control setting across diverse tasks. The results demonstrate that our framework achieves comparable or superior performance to state-of-the-art methods in error-free settings while significantly outperforming existing methods when handling a high percentage of erroneous human preferences.

Satellite communication with beam hopping is a promising approach to meeting wide-area user traffic demands under on-board resource limitation. However, there is currently a lack of theoretical models that characterize the impact of beam hopping on user uplink transmission performance. In this article, two simplified beam hopping modes for satellite communication are proposed, namely, probabilistic beam hopping and deterministic beam hopping. Based on these two modes, two Markov chain models are established, which describe the states and state transitions of users. In the first model, the user state incorporates only user queue length, while the state in the second model considers both user queue length and time slot index. For both models, we first derive their steady-state probabilities and state transition probabilities, and then the steady-state probability of successful packet transmission is solved through a numerical method. Further, we theoretically derive the explicit expressions of various user performance metrics based on the steady-state probability of successful packet transmission, such as average throughput, buffer occupancy rate, packet loss rate, and transmission delay, and simulation results verify the correctness of all expressions. By simulation, it is shown that the minimum beam illumination probability or time length can be determined given specific performance requirements to guide system design, and there is an inflection point for the improvement in system performance with the increase of beam resource. Moreover, it is found that deterministic beam hopping achieves lower user delay compared to probabilistic beam hopping when the system is near to saturation.