CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Focus

Problem 1: BLE crowdsourced offline high-precision location aggregation (基于非线性优化的众包离线高精度位置聚合). See problems.pdf page 1 for the full problem statement.

Current phase: Project proposal (项目申请书) — research and writing, not implementation.

Core optimization problem: $$\hat{x} = \argmin_x \sum_i \omega_i (f(r_i) - |x - I_i|_2)^2$$

Targets: Static CEP90 ≤ 30m (≥10 samaritans); Moving ≥20% improvement over baseline.

Phase Context

This is a fresh start focused on literature survey, method selection, and proposal writing. The archived-ex-research-problem1/ directory contains earlier exploratory work — treat as background context only, not as the proposed approach. The proposal should demonstrate independent research depth and may adopt entirely different methods.

Research Documents

• docs/questions.md — 待确认问题集（申请书规范、数据、baseline、专利、BLE 技术边界、实时性需求）
• docs/literature-deep-dive.md — 31 篇论文深度调研汇总（锚点不确定性、RSSI 物理极限、设备异构、滤波、权重设计、室外场景、Apple Find My）
• docs/method-candidate-*.md — 方法候选文档：
  • weight-learning — 数据驱动的 $\omega_i$ 学习（参数化/bilevel/EM）
  • filtering — 卡尔曼滤波 vs 粒子滤波对比
  • map-constraint — 环境与地图约束融合（路网、建筑、交通模式先验、立体空间修正）
  • map-aided-filtering — 路网约束粒子滤波（融合 filtering + map-constraint 的统一方案）

Literature Corpus

literauture-research/challenge1_literature/ 包含 31+ 篇论文 PDF/HTML 及配套文件。核心发现见 docs/literature-deep-dive.md。最高价值论文：R3-R6（锚点不确定性）、B13（设备异构）、B4/S3（RSSI 物理极限）、B7（降级策略）。

Conventions

• Research documents and proposal drafts: docs/
• Method candidates: docs/method-candidate-<name>.md
• All source code in English; documents may be in Chinese
• Documents use YAML frontmatter (tags, model, created)