# Can AI Understand the Language of Origami?

arXiv: https://arxiv.org/abs/2603.13856 · HTML: https://mlsanity.com/paper/2603.13856/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 26 references that are in the map.
- First posted on arXiv 14 Mar 2026.

## Builds on (most cited first, 16 of 16)

- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives](https://mlsanity.com/paper/2311.18259/) (2023)
- [ProcTHOR: Large-Scale Embodied AI Using Procedural Generation](https://mlsanity.com/paper/2206.06994/) (2022)
- [Physion: Evaluating Physical Prediction from Vision in Humans and Machines](https://mlsanity.com/paper/2106.08261/) (2021)
- [Infinite Photorealistic Worlds using Procedural Generation](https://mlsanity.com/paper/2306.09310/) (2023)
- [Bongard-LOGO: A New Benchmark for Human-Level Concept Learning and Reasoning](https://mlsanity.com/paper/2010.00763/) (2020)
- [CaptainCook4D: A Dataset for Understanding Errors in Procedural Activities](https://mlsanity.com/paper/2312.14556/) (2023)
- [D2CSG: Unsupervised Learning of Compact CSG Trees with Dual Complements and Dropouts](https://mlsanity.com/paper/2301.11497/) (2023)
- [Are Deep Neural Networks SMARTer than Second Graders?](https://mlsanity.com/paper/2212.09993/) (2022)
- [VLMaterial: Procedural Material Generation with Large Vision-Language Models](https://mlsanity.com/paper/2501.18623/) (2025)
- [PhyBlock: A Progressive Benchmark for Physical Understanding and Planning via 3D Block Assembly](https://mlsanity.com/paper/2506.08708/) (2025)
- [ORIGAMISPACE: Benchmarking Multimodal LLMs in Multi-Step Spatial Reasoning with Mathematical Constraints](https://mlsanity.com/paper/2511.18450/) (2025)
- [ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points](https://mlsanity.com/paper/2503.02745/) (2025)
- [PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction](https://mlsanity.com/paper/2404.10620/) (2024)
- [Learn2Fold: Structured Origami Generation with World Model Planning](https://mlsanity.com/paper/2603.29585/) (2026)
- [GamiBench: Evaluating Spatial Reasoning and 2D-to-3D Planning Capabilities of MLLMs with Origami Folding Tasks](https://mlsanity.com/paper/2512.22207/) (2025)

## Built on by (most cited first, 4 of 4)

- [Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias](https://mlsanity.com/paper/2607.11871/) (2026)
- [COrigami: An AI Pipeline for Co-Designing Flat-Foldable Visually Recognisable Origami](https://mlsanity.com/paper/2606.26299/) (2026)
- [FoldingAgent: Inferring Parametric Origami Procedures from Demonstration Videos](https://mlsanity.com/paper/2609.00377/) (2026)
- [GTA: Graph Theory Agent and Benchmark for Algorithmic Graph Reasoning with LLMs](https://mlsanity.com/paper/2609.12265/) (2026)

## Abstract (arXiv)

Building AI systems that can plan, act, and create in the physical world requires more than pattern recognition. Such systems must reason about the generative mechanisms and constraints governing physical processes, using structured representations that connect observations, actions, and their effects. Yet, many existing benchmarks study these capabilities separately, focusing either on visual recognition or on abstract symbolic or programmatic reasoning. Origami provides a natural testbed that integrates these abilities: constructing shapes through folds requires visual perception, reasoning about geometric and physical constraints, and sequential planning, while remaining sufficiently structured for systematic evaluation. We introduce OrigamiBench, a benchmark for evaluating programmatic understanding of the mechanisms underlying origami synthesis through a high-level language of physically grounded fold actions. Experiments with modern vision-language models reveal that scaling model size alone does not reliably improve reasoning about physical transformations. Moreover, models struggle to ground programmatic information in visual observations, suggesting that visual and language representations remain weakly integrated.