Notebook 3: Reinforcement Fields in GRL¶
This directory contains the main notebook and supplementary materials for understanding GRL's reinforcement fields.
Contents¶
| File | Description |
|---|---|
03_reinforcement_fields.ipynb |
Main notebook: 2D navigation domain, particles, Q⁺ field, policy inference |
03a_particle_coverage_effects.ipynb |
Supplementary: Visual proof of how particle coverage affects policy field behavior |
particle_vs_gradient_fields.md |
Theory note: Detailed comparison of particle-based Q⁺ vs true gradient fields |
Key Question Addressed¶
Why do policy arrows appear parallel instead of converging on the goal?
The supplementary notebook (03a) provides visual proof that:
- Sparse particles → parallel arrows (limited directional info)
- Rich particles → converging arrows (diverse directional info)
- True gradient → perfect convergence (geometric, not learned)
Related¶
- Field series overview:
../README.md - Theory docs:
../../../docs/theory/particle_vs_gradient_fields.md(canonical location)