vln-keynews-weekly

name	vln-keynews-weekly
description	Find and analyze key embodied navigation (VLN/VLA/WAM) projects and models from the past week in the user's collected repository. Trigger this skill when the user asks to analyze, summarize, or extract key projects, models, or weekly updates from the ENav repository (e.g., '分析 ENav 里的资料', '提取上一周的具身导航关键项目', '每周具身智能导航项目总结').

VLN Key News Weekly Analyzer

This skill automates the scanning, extraction, and evaluation of weekly key projects/models in the field of embodied navigation (VLN, VLA, WAM) from WeChat official account articles and Perplexity summaries.

Core Workflow

Step 1: Date Range Determination

Determine the date range of "last week" (Monday to Sunday of the previous week) based on the current local time. For example, if the current date is 2026-06-09, "last week" is 2026-06-01 to 2026-06-07.

Step 2: Run Extraction Script

Run the bundled analysis script to scan the target directories (default items directory is G:\其他计算机\我的主机\Repository\ENav\items).

Run command in host PowerShell:

$env:PYTHONUTF8=1; python [SkillPath]/scripts/find_weekly_keynews.py --date [CurrentDateYMD] --output [ScratchPath]/weekly_analysis.txt

Note: Replace [SkillPath] with the absolute path of this skill folder (e.g., C:/Users/tingd/.gemini/config/skills/vln-keynews-weekly) and [ScratchPath] with the conversation's scratch directory.

Step 3: Analyze & Filter Projects

Read the output text file generated by the script. Apply the following filters to select the projects to recommend:

1. Target Domain: Focus strictly on the Vision-Language Navigation (VLN / VLN-CE) field.
2. Exclude Drones/UAVs: Exclude any drone or unmanned aerial vehicle (UAV) related navigation, tracking, or retrieval projects.
3. R2R-CE SR Threshold: Identify the R2R-CE Success Rate (SR) for each candidate project. Do not recommend projects with an R2R-CE SR below 50%, unless they feature major architectural innovations (e.g., zero-shot SOTA, panoramic-free/ego-only designs, novel abstract maps, or rule-compliance) that make them exceptionally worthy of review.
4. Architectural Innovation: Prioritize works with novel designs (e.g., decoupled decision translation, uncertainty-aware reasoning, active exploration, abstract obstacle maps).
5. Open-Source Code: Prioritize projects that have active GitHub repositories with released code.

Step 4: Create Report Artifact

Write a structured report as a user-facing markdown artifact named weekly_nav_analysis.md in the conversation's artifact directory. The report must contain:

1. Summary Table: Compare the selected projects across Title, Org/Authors, R2R-CE SR, Architectural Innovation, and GitHub link.
2. Detailed Analysis per Project: Analyze each project including:
   • Basic Info (Paper/WeChat links, Authors)
   • R2R-CE SR & Core Performance (explicitly list the R2R-CE SR, or specify if not evaluated on standard R2R-CE)
   • Architectural Innovations (design breakdown)
   • Open Source repository link
3. Research Trends Summary: Highlight the general direction of last week's publications (e.g., zero-shot VLN-CE, active exploration reinforcement learning, abstract spatial mapping, rule-compliant navigation).
4. Clickable Links: Ensure all local files and GitHub repositories are fully clickable using the file:// scheme with forward slashes for Windows.

Step 5: Final Response

Point the user to the newly created report artifact and write a concise, summary-level response. Directly list the R2R-CE SR of each recommended project in the response text. Do not duplicate the entire analysis in the response text.

---

Bundled Resources

scripts/find_weekly_keynews.py

A python script that calculates target date ranges, scans matching markdown files in the ENav repository, parses metadata, and writes a preliminary analysis output to a UTF-8 text file.