# M:3L — Mathematical Modeling & Machine Learning Laboratory

> M:3L (also written M3L or M:3L Lab) is a hybrid research and engineering laboratory established in June 2025 at the Institute of Mechanics of the National Academy of Sciences of the Republic of Armenia in Yerevan, Armenia.

For a comprehensive machine-readable description of M:3L's research, projects, benchmarks, capabilities, publications, technologies, applications, and collaboration areas, see [llms-full.txt](https://m3l.am/llms-full.txt).

M:3L develops 3D and CAD intelligence that connects multi-view or unposed imagery and point clouds to zero-shot 3D understanding, open-vocabulary object detection, scene layouts, editable parametric CAD, and executable CAD programs. Its broader work covers agentic AI, scientific and physics-driven machine learning, differentiable FEM, computational mechanics, inverse design, metamaterials, wave control, and event-sequence models.

Important evidence includes Z3D on ScanRefer and Nr3D; Zoo3D on ScanNet200 and ARKitScenes; TUN3D on ScanNet, S3DIS, ARKitScenes, and Structured3D; and cadrille on DeepCAD, Fusion360, CC3D, and Omni-CAD. Detailed paper-reported metrics and protocol boundaries are in [llms-full.txt](https://m3l.am/llms-full.txt) and the publication records below.

## Core laboratory information

- [Comprehensive M:3L guide](https://m3l.am/llms-full.txt): Self-contained research, project, benchmark, technology, application, and collaboration evidence.
- [M:3L homepage](https://m3l.am/index.md): Compact overview of the laboratory and website.
- [About M:3L](https://m3l.am/about.md): Identity, history, institutional home, location, specialization, and collaboration profile.
- [M:3L Team](https://m3l.am/team.md): Researchers, roles, affiliations, supported expertise, profiles, and selected project relationships.
- [Contact M:3L](https://m3l.am/contact.md): Contact information and collaboration areas.

## Research map

- [Research overview](https://m3l.am/research.md): How M:3L's AI, spatial, engineering, scientific-ML, and computational-mechanics directions interact.
- [3D Intelligence](https://m3l.am/research/3d-intelligence.md): Zero-shot 3D grounding, open-vocabulary detection, unposed imagery, scene layouts, point clouds, spatial intelligence, and benchmark evidence.
- [CAD Intelligence](https://m3l.am/research/cad-intelligence.md): Multimodal CAD reconstruction, executable CadQuery programs, parametric geometry, online RL, validity, and CAD benchmark evidence.
- [Agentic AI](https://m3l.am/research/agentic-ai.md): Tool-using and multi-agent systems, RAG, MCP, memory, scientific agents, and CAD/CAE/simulation orchestration.
- [Physics-Driven Machine Learning](https://m3l.am/research/physics-driven-ml.md): Scientific ML, governing physics, neural fields, PINNs, graph neural networks, surrogates, and inverse problems.
- [Differentiable Finite Element Methods](https://m3l.am/research/differentiable-fem.md): Automatic differentiation through FEM, gradient-based inverse design, and simulation-to-optimization loops.
- [Metamaterials and Wave Control](https://m3l.am/research/metamaterials.md): Elastic metamaterials, Rayleigh-wave control, elastic cloaking, chiral structures, microstructure design, and FEM-based optimization.
- [Event Sequence and Temporal Models](https://m3l.am/research/event-sequence-models.md): Semantic and temporal event embeddings, outcome-aware representations, retrieval, monitoring, and prediction.

## 3D and CAD publications

- [Z3D: Zero-Shot 3D Visual Grounding from Images](https://m3l.am/publications/z3d.md): ACL 2026; methods, ScanRefer/Nr3D evidence, authors, arXiv, DOI, and code.
- [Zoo3D: Zero-Shot 3D Object Detection at Scene Level](https://m3l.am/publications/zoo3d.md): CVPR 2026; training-free and self-supervised detection, ScanNet/ARKitScenes evidence, authors, arXiv, DOI, and code.
- [TUN3D: Towards Real-World Scene Understanding from Unposed Images](https://m3l.am/publications/tun3d.md): ICRA 2026; joint layout and object understanding without known poses or external depth, benchmark evidence, project page, and code.
- [cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning](https://m3l.am/publications/cadrille.md): ICLR 2026 Oral; point-cloud/image/text-to-CAD, online RL, DeepCAD/Fusion360/CC3D evidence, and code.
- [All M:3L publications](https://m3l.am/publications.md): Authoritative records with status, venue, authors, methodology, datasets, results, metrics, links, and evidence boundaries.

## Projects and activity

- [News & Events](https://m3l.am/news.md): Research projects, collaborations, grants, publication announcements, conferences, and laboratory activity.

## Machine-readable resources

- [Comprehensive M:3L knowledge document](https://m3l.am/llms-full.txt): Detailed self-contained evidence for AI agents and research systems.
- [Machine-readable site index](https://m3l.am/api/site-index.json): JSON resource graph with canonical URLs, Markdown URLs, categories, descriptions, and research topics.
- [XML sitemap](https://m3l.am/sitemap.xml): Canonical human-readable pages.

## External and institutional references

- [M:3L on LinkedIn](https://www.linkedin.com/company/mathematical-modeling-and-machine-learning-lab-m3l/posts/): Public research announcements and laboratory activity.
- [Institute of Mechanics, NAS RA](http://www.mechins.sci.am/): Institutional home of M:3L.

## Optional

- [M:3L human-readable homepage](https://m3l.am/): Main interactive website for human visitors.