Search papers, labs, and topics across Lattice.
This paper introduces RuleMaze, a benchmark designed to evaluate the ability of multimodal large language models (MLLMs) to perform visual spatial planning while adhering to natural-language rules. The authors propose a novel method called Disentangled Multimodal Planning (DMP), which separates perception, execution, and rule verification to enhance rule compliance and generalization to complex, unseen rules. Experimental results show that DMP significantly outperforms traditional end-to-end planning approaches, highlighting the importance of interpretable reasoning in MLLMs' spatial reasoning capabilities.
RuleMaze reveals that separating perception, execution, and rule verification can dramatically enhance MLLMs' ability to follow complex natural-language instructions in spatial planning tasks.
Multimodal large language models (MLLMs) combine linguistic reasoning with visual perception, yet their ability to perform visual spatial planning under explicit or previously unseen rule constraints remains underexplored. This setting requires models to jointly understand spatial layouts, interpret natural-language rules, and plan valid actions accordingly. To address this gap, we introduce RuleMaze, a controllable benchmark in which MLLMs must navigate mazes while obeying natural-language rules of varying complexity. RuleMaze isolates rule-compliant spatial planning by requiring accurate perception, rule interpretation, and constrained action planning. To enable scalable and systematic rule construction, we propose Language-Logic-Function Hybridization, which automatically generates natural-language rules and translates them into logical representations and executable validators, eliminating manual rule engineering. To improve rule following and generalization, we introduce Disentangled Multimodal Planning (DMP), which separates perception, execution, and rule verification through interpretable reasoning primitives. By disentangling these components, DMP facilitates systematic generalization to more complex and previously unseen rules, while providing transparent intermediate planning traces. Experiments demonstrate that DMP substantially improves rule compliance and planning success compared to end-to-end textual planning baselines. Overall, RuleMaze establishes a principled benchmark for studying grounded and interpretable rule-based spatial planning in MLLMs. Code is available at https://github.com/oceanflowlab/RuleMaze.
Tsinghua AI