LLMs for Text-Based Exploration and Navigation Under Partial Observability
Stephan Sandfuchs, Maximilian Melchert, J\"org Frochte
Feedback
Why It Matters
Understanding how LLMs can be effectively utilized in navigation and exploration tasks under partial observability has significant implications for applications in logistics, search-and-rescue, and robotics.
Contributions
- Introduction of a reproducible benchmark for evaluating LLMs in exploration and navigation tasks under partial observability.
Insights
- Reasoning-tuned models outperform others in navigation tasks but struggle with efficiency compared to oracle paths.
Limitations
- The reliance on a fixed ASCII gridworld may limit the generalizability of the findings to more complex real-world scenarios.
Tags
- agent
- benchmark
- llm
- reasoning
Abstract
arXiv:2604.09604v1 Announce Type: new Abstract: Exploration and goal-directed navigation in unknown layouts are central to inspection, logistics, and search-and-rescue. We ask whether large language models (LLMs) can function as \emph{text-only} controllers under partial observability -- without code execution, tools, or program synthesis. We introduce a reproducible benchmark with oracle localisation in fixed ASCII gridworlds: each step reveals only a local $5\times5$ window around the agent and the model must select one of \texttt{UP/RIGHT/DOWN/LEFT}. Nine contemporary LLMs ranging from open/proprietary, dense / Mixture of Experts and instruction- vs. reasoning-tuned are evaluated on two tasks across three layouts of increasing difficulty: \emph{Exploration} (maximising revealed cells) and \emph{Navigation} (reach the goal on the shortest path). The experimental results are evaluated on quantitative metrics including \emph{success rate}, \emph{efficiency} such as normalised coverage and \emph{path length} vs. oracle as well as qualitative analysis. Reasoning-tuned models reliably complete navigation across all layouts, yet remain less efficient than oracle paths. Few-shot demonstrations in the prompt chiefly help these Reasoning-tuned models by reducing invalid moves and shortening paths, while classic dense instruction models remain inconsistent. We observe characteristic action priors (UP/RIGHT) that can induce looping under partial observability. Overall, training regimen and test-time deliberation predict control ability better than raw parameter count. These findings suggest lightweight hybridisation with classical online planners as a practical route to deployable partial map systems.