Embodied agents tasked with complex scenarios, whether in real or simulated environments, rely heavily on robust planning capabilities. When instructions are formulated in natural language, large language models (LLMs) equipped with extensive linguistic knowledge can play this role. However, to effectively exploit the ability of such models to handle linguistic ambiguity, to retrieve information from the environment, and to ground plans in the available skills of an agent, an appropriate architecture must be designed. We propose a Hierarchical Embodied Language Planner, called HELP, consisting of a set of LLM-based agents, each dedicated to solving a different subtask. A high-level planner resolves ambiguous instructions by querying the environment and decomposes them into unambiguous natural language sub-tasks, while a low-level planner translates each sub-task into an executable sequence of actions grounded in the agent’s skills. We evaluate HELP on a household task, validating it on both a dataset of ambiguous references and the ALFRED benchmark, and perform real-world experiments with an embodied agent equipped with a robotic arm. We also focus on the use of open source LLMs with a relatively small number of parameters, enabling autonomous on-device deployment without reliance on proprietary APIs.
DOI: 10.1007/978-3-032-34387-1_5
Скачать сборник (PDF) с сайта Springer Nature (англ.): https://link.springer.com/content/pdf/10.1007/978-3-032-34387-1.pdf
Скачать статью (PDF) или читать онлайн на arXiv.org (англ.): https://arxiv.org/html/2512.21723
ResearchGate: https://www.researchgate.net/publication/399134687_HELP_Hierarchical_Embodied_Language_Planner_for_Household_Tasks
Korchemnyi, A., Onishchenko, A., Bakaeva, E., Kovalev, A., Panov, A. (2027). HELP: Hierarchical Embodied Language Planner for Household Tasks // In: Ronzhin, A., Gribova, V., Meshcheryakov, R. (eds) Interactive Collaborative Robotics. ICR 2026. Lecture Notes in Computer Science, vol 16790, pp. 62–77.