Reproduce “HELP: Hierarchical Embodied Language Planner for Household Tasks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alexandr V. Korchemnyi, Anatoly O. Onishchenko, Eva A. Bakaeva, Alexey K. Kovalev, Aleksandr I. Panov
The Hierarchical Embodied Language Planner (HELP) framework enables robots to perform complex household tasks from natural language instructions by employing a two-stage planning pipeline with medium-sized open-source LLMs and integrating environmental feedback. The system successfully completed 80% of real-world tasks and maintained robust performance on long-horizon plans, contrasting with the significant degradation observed in baseline methods.
Key insights
- Hierarchical task decomposition is vital for robust and coherent planning of long-horizon tasks, significantly outperforming monolithic LLM approaches.
- Dynamically selecting relevant in-context examples based on semantic similarity substantially improves planning quality across various LLM sizes.
- Real-time environmental feedback and perceptual grounding are critical for resolving ambiguities and ensuring generated plans are precisely executable in the physical world.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.