Uncertainty Mitigation and Intent Inference: A Dual-Mode Human-Machine Joint Planning System

TL;DR

A human-robot joint planning system that tackles two sources of uncertainty at once: task-relevant knowledge gaps (resolved through two-way conversation) and latent human intent (inferred without any explicit communication).

Key contributions:

• Uncertainty-mitigation mode: LLM-assisted active elicitation with hypothesis-augmented A* search and a dynamic-programming querying policy — cutting interaction cost by 51.9%.
• Intent-aware mode: a probabilistic belief over the human’s latent task intent from spatial and directional cues, enabling coordination-aware task selection — reducing task execution time by 25.4%.
• Validated in Gazebo simulation and real-world UAV deployments with a VLM-based 3D semantic perception pipeline.

BibTeX

@misc{fang2026uncertainty,
      title={Uncertainty Mitigation and Intent Inference: A Dual-Mode Human-Machine Joint Planning System},
      author={Zeyu Fang and Yuxin Lin and Cheng Liu and Beomyeol Yu and Zeyuan Yang and Rongqian Chen and Taeyoung Lee and Mahdi Imani and Tian Lan},
      year={2026},
      eprint={2603.07822},
      archivePrefix={arXiv},
      url={https://arxiv.org/abs/2603.07822},
}