CubeBench: Diagnosing Interactive, Long-Horizon Spatial Reasoning Under Partial Observations
This addresses the problem of enabling LLM agents to handle interactive, long-horizon spatial reasoning under partial observations, which is crucial for physical-world applications, but it is incremental as it focuses on diagnosing limitations rather than solving them.
The paper tackled the challenge of LLM agents forming spatial mental models for physical-world deployment by introducing CubeBench, a benchmark based on the Rubik's Cube, and found that leading LLMs had a 0.00% pass rate on long-horizon tasks, exposing failures in long-term planning.
Large Language Model (LLM) agents, while proficient in the digital realm, face a significant gap in physical-world deployment due to the challenge of forming and maintaining a robust spatial mental model. We identify three core cognitive challenges hindering this transition: spatial reasoning, long-horizon state tracking via mental simulation, and active exploration under partial observation. To isolate and evaluate these faculties, we introduce CubeBench, a novel generative benchmark centered on the Rubik's Cube. CubeBench uses a three-tiered diagnostic framework that progressively assesses agent capabilities, from foundational state tracking with full symbolic information to active exploration with only partial visual data. Our experiments on leading LLMs reveal critical limitations, including a uniform 0.00% pass rate on all long-horizon tasks, exposing a fundamental failure in long-term planning. We also propose a diagnostic framework to isolate these cognitive bottlenecks by providing external solver tools. By analyzing the failure modes, we provide key insights to guide the development of more physically-grounded intelligent agents.