PuzzleWorld: A Benchmark for Multimodal, Open-Ended Reasoning in Puzzlehunts

Puzzlehunts are a genre of complex, multi-step puzzles lacking well-defined problem definitions. In contrast to conventional reasoning benchmarks consisting of tasks with clear instructions, puzzlehunts require models to discover the underlying problem structure from multimodal evidence and iterative reasoning, mirroring real-world domains such as scientific discovery, exploratory data analysis, or investigative problem-solving. Despite recent progress in foundation models, their performance on such open-ended settings remains largely untested. In this paper, we introduce PuzzleWorld, a large-scale benchmark of 667 puzzlehunt-style problems designed to assess step-by-step, open-ended, and creative multimodal reasoning. Each puzzle is annotated with the final solution, detailed reasoning traces, and cognitive skill labels, enabling holistic benchmarking and fine-grained diagnostic analysis. Most state-of-the-art models achieve only 1-2% final answer accuracy, with the best model solving only 14% of puzzles and reaching 40% stepwise accuracy. To demonstrate the value of our reasoning annotations, we show that fine-tuning a small model on reasoning traces improves stepwise reasoning from 4% to 11%, while training on final answers alone degrades performance to near zero. Our error analysis reveals that current models exhibit myopic reasoning, are bottlenecked by the limitations of language-based inference, and lack sketching capabilities crucial for visual and spatial reasoning. We release PuzzleWorld at https://github.com/MIT-MI/PuzzleWorld to support future work on building more general, open-ended, and creative reasoning systems.

Swarm Algorithms for Dynamic Task Allocation in Unknown Environments

Robot swarms, systems of many robots that operate in a distributed fashion, have many applications in areas such as search-and-rescue, natural disaster response, and self-assembly. Several of these applications can be abstracted to the general problem of task allocation in an environment, in which robots must assign themselves to and complete tasks. While several algorithms for task allocation have been proposed, most of them assume either prior knowledge of task locations or a static set of tasks. Operating under a discrete general model where tasks dynamically appear in unknown locations, we present three new swarm algorithms for task allocation. We demonstrate that when tasks appear slowly, our variant of a distributed algorithm based on propagating task information completes tasks more efficiently than a Levy random walk algorithm, which is a strategy used by many organisms in nature to efficiently search an environment. We also propose a division of labor algorithm where some agents are using our algorithm based on propagating task information while the remaining agents are using the Levy random walk algorithm. Finally, we introduce a hybrid algorithm where each agent dynamically switches between using propagated task information and following a Levy random walk. We show that our division of labor and hybrid algorithms can perform better than both our algorithm based on propagated task information and the Levy walk algorithm, especially at low and medium task rates. When tasks appear fast, we observe the Levy random walk strategy performs as well or better when compared to these novel approaches. Our work demonstrates the relative performance of these algorithms on a variety of task rates and also provide insight into optimizing our algorithms based on environment parameters.

Hierarchical Grammar-Induced Geometry for Data-Efficient Molecular Property Prediction

The prediction of molecular properties is a crucial task in the field of material and drug discovery. The potential benefits of using deep learning techniques are reflected in the wealth of recent literature. Still, these techniques are faced with a common challenge in practice: Labeled data are limited by the cost of manual extraction from literature and laborious experimentation. In this work, we propose a data-efficient property predictor by utilizing a learnable hierarchical molecular grammar that can generate molecules from grammar production rules. Such a grammar induces an explicit geometry of the space of molecular graphs, which provides an informative prior on molecular structural similarity. The property prediction is performed using graph neural diffusion over the grammar-induced geometry. On both small and large datasets, our evaluation shows that this approach outperforms a wide spectrum of baselines, including supervised and pre-trained graph neural networks. We include a detailed ablation study and further analysis of our solution, showing its effectiveness in cases with extremely limited data. Code is available at https://github.com/gmh14/Geo-DEG.