Spaces:
Sleeping
Sleeping
Origami RL Environment — Research Index
What We're Building
An OpenEnv environment where an LLM learns to design optimal origami fold patterns — solar panel packing, deployable structures, medical stents. LLM generates a fold_strategy() function (code-as-policy), executed against a bar-and-hinge physics simulation.
Architecture (START HERE)
| File | What's In It |
|---|---|
| plan/architecture.md | Full architecture: action space, state, physics, rewards, rendering, project structure, implementation order |
| plan/openenv_arch.md | Complete OpenEnv environment: repo structure, Pydantic models, engine (paper/fold/physics/validation/metrics/materials), renderer (2D/3D/screenshots/GIF recording/export), environment class, React frontend, app+Docker, client, task system, API reference, deployment, testing |
Decisions (Locked)
| Decision | Choice |
|---|---|
| LLM interaction | Code-as-policy (LLM writes fold_strategy() function) |
| Action space | Named fold ops (valley/mountain + fold line + angle) |
| State format | FOLD-compatible JSON |
| Physics engine | Bar-and-hinge model (NumPy port of Ghassaei) |
| Validation | Kawasaki + Maekawa + triangle-triangle intersection |
| Primary task | Solar panel packing (Miura-ori discovery) |
| Training render | matplotlib headless |
| Demo render | React + @react-three/fiber |
| Training | GRPO via TRL + Unsloth on Colab |
| Deployment | Docker Space on HF Spaces |
OpenEnv (The Framework)
| File | What's In It |
|---|---|
| openenv/overview.md | OpenEnv architecture, API, types, project structure, deployment |
| openenv/2048_pattern.md | Code-as-policy pattern, reward functions, GRPO training |
| openenv/2048_example.py | Full extracted code from Unsloth 2048 Colab (636 lines) |
Origami Domain Knowledge
Quick Reference
| File | What's In It |
|---|---|
| origami/fold_types_deep.md | All fold operations, Huzita-Justin axioms, crane step-by-step (31 steps), compression patterns (Miura-ori, Kresling, flasher), complexity scaling |
| origami/math_physics_deep.md | Kawasaki/Maekawa theorems with code, bar-and-hinge model, energy formulas, strain computation, rigid foldability, computational complexity table |
| origami/rendering_research.md | Rendering options: Ghassaei simulator, OrigamiOdyssey (R3F), Three.js in React, Gradio integration, recording |
| origami/applications_deep.md | Real-world apps: NASA solar panels, JWST, stents, self-folding robots, metamaterials |
Earlier Research (Summaries)
| File | What's In It |
|---|---|
| origami/simulation_engines.md | Ghassaei, rigid-origami Gym env, SWOMPS, Tachi |
| origami/fold_format.md | FOLD file format — JSON standard for crease patterns |
| origami/physics.md | Physics summary (Kawasaki, Maekawa, simulation approaches) |
| origami/materials.md | Material properties (paper, mylar, aluminum), stress viz |
| origami/metrics.md | All metrics: validity, compactness, stress, shape similarity |
| origami/existing_work.md | Prior work: IJCAI 2023, Nature 2022, UCLA robotics |
| origami/python_tools.md | Libraries: rigid-origami, PyOri, numpy, trimesh |
Deliverables Checklist
- Engine: paper.py, fold_engine.py, physics.py, validation.py, metrics.py
- OpenEnv server: models.py, origami_environment.py, app.py, Dockerfile
- Reward functions: code_valid, physically_valid, fold_quality
- Training notebook: Colab with GRPO + Unsloth/TRL
- Rendering: matplotlib (training) + React/R3F (demo)
- Deploy to HF Spaces
- 1-minute demo video on YouTube
- Public GitHub repo