@inproceedings{zheng-etal-2026-pilot,
    title = "{PILOT}: Planning via Internalized Latent Optimization Trajectories for Large Language Models",
    author = "Zheng, Haoyu  and
      Zhu, Yun  and
      Yuan, Yuqian  and
      Yuan, Bo  and
      Zhang, Wenqiao  and
      Tang, Siliang  and
      Xiao, Jun",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.acl-long.236/",
    pages = "5204--5222",
    ISBN = "979-8-89176-390-6",
    abstract = "Strategic planning is critical for multi-step reasoning, yet compact Language Language Models (LLMs) often lack the capacity to formulate global strategies, leading to error propagation in long-horizon tasks. Our analysis reveals that LLMs possess latent reasoning capabilities that can be unlocked when conditioned on explicit plans from a teacher model; however, runtime reliance on external guidance is often impractical due to latency and availability constraints. To bridge this gap, we propose \textbf{PILOT} (\textbf{P}lanning via \textbf{I}nternalized \textbf{L}atent \textbf{O}ptimization \textbf{T}rajectories), a non-invasive framework designed to internalize the strategic oversight of large models into intrinsic \textit{Latent Guidance}. Instead of altering backbone weights, PILOT employs a lightweight Hyper-Network to synthesize a query-conditioned \textit{Latent Guidance}. This vector acts as an internal steering mechanism, guiding the model{'}s representations toward optimal reasoning paths. Extensive experiments on mathematical and coding benchmarks demonstrate that PILOT effectively stabilizes reasoning trajectories, consistently outperforming strong baselines (e.g., +8.9{\%} on MATH500) with negligible inference latency. Our code is available at: \url{https://anonymous.4open.science/r/PILOT-B266}"
}