06 — 開發故事Development Story

從一個問題,
做到一套系統。

From one problem
to a whole system.

迭代過程、走過的彎路,以及為什麼最後長成現在這樣。

How the project actually evolved — the detours, the dead ends, and why we ended up here.

源自尖石鄉與泰雅文化的啟發INSPIRED BY JIANSHI & ATAYAL CULTURE

起點

Origin

團隊中有兩位成員在新竹縣尖石鄉長大,並在當地就讀小學。從小生活在陡峭山勢、 深邃溪谷與豐富的泰雅文化之中,讓他們與這片土地建立了深厚的連結。然而,這片美麗的山林 同時也容易受到地震、颱風、土石流與通訊中斷的威脅。受到家鄉及當地真實防災需求的啟發, 我們以尖石鄉為原型設計搜救情境模型,期盼 LinkGuard 能守護這裡的人們,以及他們所珍愛的山林。

Two of our team members grew up in Jianshi Township, Hsinchu County, and attended elementary school there. Surrounded by steep mountains, deep valleys, and the rich Atayal culture, they developed a deep connection with the local environment from an early age. However, this beautiful landscape is also vulnerable to earthquakes, typhoons, landslides, and communication disruptions. Inspired by their hometown and its real disaster challenges, we designed our rescue scenario model based on Jianshi Township, hoping that LinkGuard can help protect both the people and the mountains they cherish.

場域輪廓Landscape

527.6 km² 的山林鄉鎮

A 527.6 km² mountain township

尖石位於新竹縣東南方,海拔約 200–3,000 公尺。山勢、溪谷與分散聚落,使災時通訊與抵達現場都更困難。

Jianshi lies in southeastern Hsinchu County, rising from roughly 200 to 3,000 metres. Mountains, valleys and dispersed settlements make disaster communication and access difficult.

前山/後山Two regions

兩個流域,兩種救援條件

Two watersheds, two rescue contexts

前山屬頭前溪流域,交通與人口較集中;後山屬大漢溪流域,山勢更高、森林更密,包含司馬庫斯、鎮西堡等部落。

The front mountains follow the Touqian River and have better access; the remote Dahan River side is higher and more forested, including Smangus and Cinsbu.

文化核心Cultural heart

泰雅族與山林共生

Atayal life with the mountains

織布、狩獵、農耕、河川漁撈與祖靈信仰延續至今;司馬庫斯、鎮西堡、那羅與馬里光各自保存並發展部落文化。

Weaving, hunting, farming, river fishing and ancestral traditions continue today across communities including Smangus, Cinsbu, Naluo and Mrqwang.

災害風險Hazards

豪雨會放大地形阻隔

Heavy rain amplifies isolation

尖石是頭前溪與大漢溪部分水系的源區。颱風豪雨可能引發坍方、土石流與通訊中斷,正是 LinkGuard 要處理的情境。

Headwaters of the Touqian and Dahan systems cross Jianshi. Typhoons can trigger landslides, debris flows and network outages—the conditions LinkGuard is designed for.

代表場景Landmarks

從溪谷到高山巨木

From river valleys to giant trees

司馬庫斯、鎮西堡神木群、秀巒溫泉與青蛙石天空步道,呈現尖石從前山溪谷到後山高海拔森林的多樣地貌。

Smangus, the Cinsbu giant-tree forest, Xiuluan hot springs and Frog Rock show the terrain’s range from front-country valleys to remote high forests.

設計回應Design response

不依賴網路的在地韌性

Resilience without the internet

LinkGuard 以 LoRa、自組節點與本地端 AI,在基地台失效時維持生命徵象、環境資料與搜救指令的流動。

LinkGuard combines LoRa, self-forming nodes and on-device AI to keep vital signs, environmental data and rescue commands moving when cellular service fails.

為什麼最後選 LoRaWHY LORA

選型

Trade-off

災區的共同前提是「沒有基礎建設可用」。這一條就先排除了依賴基地台的行動網路, 以及需要架設 AP 的 Wi-Fi。剩下的候選裡,LoRa 的取捨最符合需求: 傳輸距離長、功耗低、免執照頻段、單顆模組成本低 —— 代價是頻寬極小,只能傳結構化的短封包,不能傳影像。 這也決定了整套系統的分工:影像交給載具,生命徵象交給 LoRa。

A disaster area shares one premise: no infrastructure to rely on. That alone rules out mobile networks, which need base stations, and Wi-Fi, which needs an access point. Among what is left, LoRa’s trade-offs fit best: long range, low power, licence-free band, low per-module cost — at the price of very little bandwidth, enough only for short structured packets, never video. That decided the division of labour for the whole system: video goes to the vehicles, vital signs go over LoRa.

方案評估結果
衛星通訊器
如 Garmin inReach 類產品
裝置約 NT$12,000 起,另需約 NT$450–2,000 月租費;而且需要開闊的天空視野,在尖石鄉的山谷與林蔭下訊號可能不穩定。
Wi-Fi/藍牙 Mesh 實測或規格上的有效距離僅約 120 公尺,不足以覆蓋約 4 公里範圍的山區搜救任務。
星狀拓撲
單一中心節點
只要指揮中心節點故障,整個網路就會癱瘓,形成單一故障點,缺乏任務所需的容錯能力。
LinkGuard
自組 LoRa Mesh
沒有單一故障點;四次實驗測得最大通訊距離為 1.96 公里;不需月租費,也不依賴開闊的天空視野。
OptionEvaluation
Satellite messenger
e.g. Garmin inReach-type products
Devices start at approximately NT$12,000, plus a monthly fee of about NT$450–2,000. They also need a clear view of the sky, so reception may be unreliable in Jianshi Township’s valleys and beneath dense tree cover.
Wi-Fi / Bluetooth mesh Its tested or specified effective range is only about 120 metres, too short for a mountain rescue operation spanning roughly 4 kilometres.
Star topology
single central node
If the command-centre node fails, the entire network goes down. This single point of failure does not provide the fault tolerance required for rescue operations.
LinkGuard
self-built LoRa mesh
No single point of failure; a maximum communication distance of 1.96 kilometres measured across four tests; no subscription fee; and no need for a clear view of the sky.
核心元件如何選定CHOOSING THE CORE COMPONENTS

選型理由

Component choices

無線核心採用 LoRa ESP32-S3+LoRa 模組。LoRa ESP32-S3 負責讀取感測資料、 封裝短封包、更新 OLED 與處理本機控制;LoRa 模組則提供低功耗、長距離的災區通訊。 模組化設計方便快速組裝、替換與維修,也讓團隊能把時間集中在多角色流程、 封包設計與現場操作可靠度。

We chose LoRa ESP32-S3 + a LoRa module. LoRa ESP32-S3 reads sensor data, packages short packets, updates the OLED and handles local controls, while the LoRa module provides low-power, long-range disaster-area communication. The modular design is quick to assemble, replace and repair, allowing the team to focus on role-based workflows, packet design and operational reliability.

心率資料優先採用支援標準 BLE Heart Rate Service 的穿戴裝置,而不是把自製感測器 當成唯一來源。這能減少額外接線與配戴負擔;若 Garmin 裝置未連線,Victim 韌體仍會 切換至測試心率模式,使通訊、介面與警示流程可以持續運作及展示。

Heart-rate input prioritises wearables that support the standard BLE Heart Rate Service instead of depending on a custom sensor as the only source. This reduces wiring and wearing complexity. If the Garmin device is unavailable, Victim firmware falls back to a test heart-rate mode so communication, UI and alert workflows remain operational.

第一線回饋改變了設計FIELD FEEDBACK CHANGED THE DESIGN

從訪談到重構

From interview to redesign

消防人員訪談讓團隊重新檢視「功能越多越好」的原始想法。災害現場同時面臨通訊不穩、 高壓決策、人員疲勞與戴手套操作,因此介面改以大按鈕、單手操作、緊急功能固定位置 為原則,並依任務角色拆分 HQ 與 Field App,避免所有資訊擠在同一個畫面。

Firefighter interviews made us reconsider the original assumption that more features always make a better tool. Disaster scenes combine unstable communications, high-pressure decisions, fatigue and gloved operation. We therefore centred the interface on large controls, one-handed use and fixed emergency actions, while separating HQ and Field App views by operational role.

這份回饋也強化了 Offline First 與分層通訊的方向:即使網際網路或手機鏈路中斷, LoRa 節點仍能持續收發短封包,OLED、蜂鳴器與震動馬達則保留最直接的現場提醒。

The same feedback reinforced an offline-first, layered communication design. Even when internet or phone links fail, LoRa nodes continue exchanging short packets, while the OLED, buzzer and vibration motor preserve immediate on-device alerts.

消防隊訪談紀錄FIRE-STATION INTERVIEW

訪談怎麼做、聽到了什麼、改了什麼

How we interviewed, what we heard, what we changed

為使系統設計貼近真實災害現場需求,團隊實地走訪消防分隊進行面對面深度訪談,受訪人員涵蓋指揮層級與第一線執勤人員,兼顧戰術指揮與現場操作兩個視角。

To ground the design in real scene requirements, the team visited a fire station for in-depth face-to-face interviews covering both command-level officers and front-line crew, so tactical command and hands-on operation were both represented.

項目Item內容Detail
受訪單位Unit新竹縣某消防分隊A fire station in Hsinchu County
受訪人數Interviewees4 位:大隊長、副大隊長、小隊長、分隊長各 1 位4: battalion chief, deputy chief, squad leader and station captain
訪談日期Date2026 年 5 月 9 日(民國 115 年)9 May 2026
訪談方式Format面對面半結構式訪談Semi-structured, face to face
訪談時長Duration約 2.5 小時About 2.5 hours

災害現場的關鍵不在於功能多寡,而在於系統能否於高壓、通訊不穩與人員疲勞之環境中維持快速、直覺且可靠之操作。

What matters on scene is not how many features a system has, but whether it stays fast, intuitive and reliable under pressure, unstable communications and fatigue.

受訪大隊長Battalion chief

現場通訊常依賴 LINE 與無線電交雜使用,訊息易遺漏或順序混亂;需要一套能整合語音、文字與位置的單一平台。

On scene we juggle LINE and radio; messages get lost or arrive out of order. We need one platform that brings voice, text and location together.

受訪分隊長Station captain

指揮層級強調指令傳遞的層級清晰與可追蹤性、避免跨單位重複回報;現場執勤層級更重視單手操作、大按鈕與離線可用性,並認為過多功能反而增加操作負擔。 訪談資料以質性分析整理後,原本偏向「資料回報與通訊」的單一取向,修正為整合式搜救輔助平台:涵蓋地圖分區、ICS 指揮、子區域建立、權限管理、任務回報、SOS 警報、照片座標回傳、搜救狀態標記、AI 輔助、離線同步、安全管制與事件日誌,並新增多角色操作模式。

Commanders stressed clear, traceable command hierarchy and avoiding duplicate cross-agency reports; front-line crew valued one-handed use, large buttons and offline availability, and warned that extra features add load. After qualitative analysis, the original "report and communicate" focus grew into an integrated rescue-support platform: map zoning, ICS command, sub-areas, permissions, task reports, SOS alerts, geotagged photos, search-status markers, AI assistance, offline sync, safety control and an event log, plus role-specific operating modes.

表五 · 地圖點線面標記Table 5 · Point / line / area markers
類型Type代表意義Meaning應用方式Use
點Point單一位置Single location受困者位置、SOS、危險點Victim location, SOS, hazard
線Line路線或邊界Route or boundary道路封鎖、危險路線、撤離動線Road closure, hazardous route, evacuation path
面Area區域範圍Region倒塌區域、搜索分區、限制進入區Collapse zone, search sector, restricted area

並支援子分區建立、搜救狀態標示、危險程度分類、照片座標綁定、臨時據點與搜救熱區顯示。

Also supports sub-sectors, search-status marking, hazard levels, geotagged photos, temporary bases and search hot-spots.

表六 · ICS 五種角色介面Table 6 · Five ICS role interfaces
角色Role功能定位Function
UCC統一指揮中心與跨區域協調Unified command centre and cross-area coordination
SCC現場指揮中心:分區管理、任務派遣與資源調度Scene command: sector management, tasking and resources
TL小隊長:管理所屬區域與彙整回報Team leader: manages a sector and consolidates reports
TE前線隊員:快速回報與 SOSTeam member: quick reports and SOS
EMT醫療組:傷患分類與後送Medical: triage and evacuation

依現場操作需求調整 ICS 標準架構,志工協調歸入 Logistics 不另列;並新增指揮權限管理、安全管制介面、已讀確認、語音翻譯與多照片回報。

The standard ICS structure is adapted to scene practice (volunteer coordination sits under Logistics); command permissions, a safety-control view, read receipts, voice translation and multi-photo reports were added.

十項高壓環境設計原則:Offline First 架構、本地暫存與自動同步、大按鈕與高對比介面、單手操作、緊急功能固定位置、二次確認、模組化設計、低電量模式、完整事件日誌,以及 Wi-Fi/LoRa/BLE 多層通訊備援。

Ten design principles for high-pressure use: offline-first architecture, local caching with auto-sync, large high-contrast controls, one-handed operation, fixed emergency actions, confirmation for critical actions, modular design, low-battery mode, a complete event log, and layered Wi-Fi / LoRa / BLE fallback.

遇到的技術瓶頸與解法WHAT BROKE, AND HOW WE FIXED IT

卡關

Bottleneck

多節點共用頻道時容易發生封包碰撞,因此我們導入 LoRa 頻道活動偵測、 Binary Exponential Backoff 與 18 個搜尋頻道分流;封包 ID 去重則避免相同訊息被重複套用。 在省電方面,系統以非阻塞排程、無線電休眠及 RSSI 動態功率調整兼顧回報速度與續航。

Multiple nodes sharing one channel caused packet collisions, so we implemented LoRa channel activity detection, binary exponential backoff and traffic distribution across 18 search channels. Packet-ID deduplication prevents repeated processing, while non-blocking scheduling, radio sleep and RSSI-based power control balance responsiveness and endurance.

目前的狀態WHERE THE SYSTEM STANDS NOW

現在

Today

LinkGuard 系統已完成整合與實機驗證。Victim、Rescue 與 HQ 三種節點韌體、 LoRa 多頻通訊、BLE Gateway、SOS/MAYDAY、隊伍命令、OLED 介面、封包去重、 頻道退避及省電模式皆已完成,並整合為可操作的災害搜救通訊系統。

LinkGuard is fully integrated and verified on hardware. The Victim, Rescue and HQ firmware roles, multi-frequency LoRa communication, BLE gateway, SOS / MAYDAY, team commands, OLED interface, packet deduplication, channel backoff and power modes are complete as one operational disaster-rescue communication system.