Inspection Is Taiwan's Furthest-Along Drone Application — Because It Routed Around BVLOS

🌏 中文版

After agricultural spraying and disaster response, I had planned to combine inspection, logistics, and surveying into one piece called "three applications stuck behind the same door" — because the industry map said inspection had "steady demand, stuck on compliance," and the BVLOS piece established that Taiwan has no BVLOS framework at all.

After the research that piece no longer exists. Inspection isn't stuck in Taiwan; it is the furthest along of the three application areas, with numbers hard enough not to need any rhetoric.

Numbers first, then why it moves while logistics doesn't.

Three sets of real numbers

Bridges: the Highway Bureau, unveiled January 2026

Taiwan has roughly 22,000 road bridges; the Ministry of Transportation's Highway Bureau administers about 3,348 of them. Its Central Region Maintenance Office has worked since 2019 with Li Ming Engineering Consultants, ITRI, and Delta Electronics on a "drone AI intelligent bridge inspection system," and has now completed field inspection of 14 river-crossing and tidal-reach bridges in Miaoli, Changhua, and Nantou.

The manpower comparison is the cleanest set of numbers in this piece. Li Ming vice president Lee Kun-che:

Conventional manual inspection required at minimum 8 inspectors and 4 vehicles over 2 days. With drones and AI, one bridge now takes 1 vehicle, 1 drone, and 5 people, roughly half a day — and crucially, no traffic control at all.

Other quantified results: the AI identifies concrete cracking, spalling, efflorescence, water seepage, steel plate and bolt corrosion, and paint peeling, at a detection accuracy above 78%, trained on over 20,000 images; operating cost is about 60% of conventional bridge inspection.

Rail: Taiwan High Speed Rail, in development since 2017

THSR runs 350 km, of which 252 km (73%) is bridge and elevated structure, requiring inspection every 2 years.

Per a briefing in late 2023, project control manager Chen Wen-tse noted that inspection had always been manual, covering at most 700 metres a day. With drones, each sortie runs 20 to 25 minutes over a 2 to 3 km round trip, saving 3 to 5 times the manpower and time.

He also noted something manual work simply cannot do: piers under riverbeds or beneath highways previously had to be examined through binoculars, sometimes without a clear view. High-resolution drone imagery shows directly whether bearing pads have shifted or steel spans have cracked.

Transmission towers: Taipower and Chia Hui Power

Taiwan has nearly 20,000 transmission towers carrying over 13,000 km of line. Taipower first introduced drones in 2019; the previous method had staff carry an infrared imager on foot to beneath the line to take temperatures, then bring the records back to the office to interpret. With a small infrared imager on a drone, nobody has to go to the site — and the measurement angle is better.

The private-sector numbers are more complete. Far Eastern group's Chia Hui Power announced in October 2024 the country's first fully automated drone inspection power plant, combining a FarEasTone 5G private network with AI. General manager Chen Kuang-tai projected a 75% reduction in manpower and a 50% reduction in inspection cost, with equipment inspection compressed from 2 days to 1. The most consequential line: inspection requires no power outage.

Why it moves while logistics doesn't

Now the question. The answer is far more precise than the word "compliance."

All three of these are segmented tasks, not continuous long-range routes.

• A bridge is a fixed point: fly to it, circle it, done
• Towers run along a line but can be flown in segments, with the operator moving along
• Each THSR sortie is a 2–3 km round trip in 20–25 minutes: one work unit, not a journey

What the BVLOS piece documented about Taiwan's regime is that it blocks genuine beyond-visual-line-of-sight — the kind requiring juridical-person status, an application 15 days in advance, and a three-month permit. Segmented fixed-point inspection can be completed within visual line of sight or within observer coverage, and never has to enter that door at all.

Logistics has no such option. Last-mile delivery is by definition "fly from A to B while you can't see it." The same regulatory threshold is barely an obstacle to segmentable work and a total blockade to work that cannot be segmented.

So the industry map's "inspection is stuck on compliance" needs refining: what's blocked is continuous long-range inspection — an entire pipeline or trunk line in one flight. Segmented fixed-point inspection isn't, and every working Taiwanese example falls in the latter category.

That also explains why the benefit narratives are always "one bridge," "one line segment," "one sortie." It isn't conservatism — it is cutting the work to the granularity regulation allows, then maximizing efficiency inside that granularity.

Safety drives adoption, not cost

The cost figures are attractive (60%, half, three quarters), but after reading all the coverage, what actually pushed these programs through is that people die.

The Highway Bureau report says it most directly: conventional road bridge inspection requires staff to walk the bridge shoulder on foot, and an inspector has previously been struck and killed. The ministry's statistic:

From 2022 to 2025, roughly 454 incidents of national highway work vehicles or crash cushion trucks being struck — one every three days on average.

One every three days. That number is more persuasive than any ROI calculation, and it explains why Lee singled out "no traffic control at all" as the crucial point — no lane closure, no exposure to being hit.

The tower side is the same thing in a different shape. Chia Hui's Chen described maintenance staff as working "like Spider-Man, having to climb towers and work at height," where the conventional method means risking their lives on a de-energized line, then climbing and walking it to inspect.

Taipower's trigger was a systemic failure: on 30 July 2017 Typhoon Nesat collapsed a transmission tower serving the Hoping power plant in Hualien, and cumulative outages reached 580,000 households nationwide.

Lined up, the three applications form a spectrum of value:

Inspection sits at the midpoint of that spectrum, and the midpoint is the easiest position to sell — it has enough numbers to persuade finance and enough lives to persuade leadership. Spraying only has the first; rescue only has the second.

It is the only application that genuinely reached Layer 5

The industry map put value at Layers 3 and 5, with Layer 5 being "turning flight data into a usable conclusion." Of Taiwan's three applications, only inspection is actually doing this, and has been for years.

• Highway Bureau: over 20,000 training images behind the AI degradation analysis, generating inspection reports automatically
• THSR: a digital imagery management system that screens large photo sets, annotates anomalies, generates inspection reports automatically, and applies automated degradation analysis to flag potential problem areas early
• Chia Hui Power: all imagery and data accumulating into a maintenance history database
• Taipower: a "drone management system" built in 2021 integrating operating manuals, professional licence issuance and expiry dates, and insurance details, with an alerting function (for instance, warning before a licence's expiry), plus records of each aircraft's sortie count and flight log alongside prohibited zone boundaries and flight application procedures

Taipower's system deserves a second look because what it manages isn't data, it's compliance. Section chief Hsu Yi-feng acknowledged that before it, drone records were "managed in something like Excel form, fairly scattered."

Past a certain fleet size, licence expiry, airspace applications, and insurance status become operational risk — which is what the compliance cost from the market entry piece looks like on the enterprise side. A spraying operator can skip it (and become one of the unlicensed). Taipower cannot.

Inspection reaches Layer 5 because it repeats. The same bridge is photographed every two years, the same tower flown several times a year; a time series forms naturally, and only then is there degradation to compare. Spraying leaves nothing behind once the pass is done; rescue is a sparse event whose sample takes many years to build.

Two details that lock onto other pieces

One: Taipower's fleet was replaced, and policy did the replacing.

Per CNA in 2021, Taipower supply section chief Hsu Yi-feng explained that the utility had once operated as many as 65 drones across its 6 regional supply offices; government cybersecurity policy then ended the use of Chinese-made aircraft, and 12 US-made units were purchased instead; in 2021 it planned to expand with 44 Taiwan-made drones.

65 → 12 → 44. That is the industry map's "non-Chinese supply chain rebuild" in miniature at a single large user, and the cost is plain: the fleet briefly shrank to under a fifth of its original size.

The same theme surfaced in bridge inspection — ITRI staff emphasized at the demonstration that the aerial drones used must be "non-red supply chain" and domestically made.

Two: the high-mountain tower inspection used the model that has crashed.

In 2021 Hung Kuang University associate professor Chang Cheng-jung led a team that, working with Taipower and engineers from AVIX Technology, flew an AVIX AXH-E230 single-rotor helicopter across the Central Mountain Range, covering 46.5 km in one hour and inspecting 79 high-voltage towers, climbing from around 800 metres to over 3,000.

The AXH-E230 is exactly the model in both TTSB investigations from the crash anatomy piece. That isn't coincidence — it is one of the few large unmanned helicopters in Taiwan capable of this class of mission. The same aircraft appearing in both a "first of its kind" story and an accident report is itself a statement about the industry's maturity.

And Chang's description of the method belongs to the same terrain-driven specification family as the "no GPS dependence" in the rescue piece:

There is no wireless coverage in the high mountains and no way to monitor continuously, so the only option is "silent flight" — the drone navigates autonomously with the link down, then flies back once the mission is complete.

The same team also worked with ITRI's materials institute on a hydrogen fuel cell unmanned helicopter dropping medical supplies at the Xinda mountain hut — 8 hours on foot, 18 minutes by air.

Three judgments

1. "Blocked by regulation" requires asking which kind of work is blocked. Taiwan having no BVLOS framework is a fact, but what it blocks is unsegmentable long-range routing. Segmented fixed-point inspection routes around it, and already has. When evaluating any application described as "waiting on regulation," first ask whether it can be cut into within-line-of-sight work units — the ones that can usually aren't waiting.
2. Safety arguments move adoption further than cost arguments. One strike incident every three days, the "Spider-Man" tower climbers, 580,000 households without power — those are why budgets pass; cost savings ride along. Inspection's advantage is that it has both kinds of argument, which is exactly what spraying and rescue each half-lack.
3. Repetition is the precondition for reaching Layer 5. Inspection grows databases and AI models (20,000 training images, automated degradation analysis, maintenance histories) because the same object gets photographed again and again. To judge whether a drone application has a Layer 5 at all, one question suffices: will it photograph the same thing a second time?

References

Bridges and rail

• CNA — Highway Bureau deploys AI vehicles and drone bridge inspection to reduce inspector risk (in Chinese; 3,348 administered bridges, 14 field inspections, 78% detection accuracy, cost at 60%, 8 people/4 vehicles/2 days versus 5 people/1 vehicle/half a day, 454 highway strike incidents)
• Liberty Times — AI drone and bridge inspection vehicle unveiled (in Chinese; same demonstration, 20,000 training images, the degradation types the AI identifies, ITRI on the non-red supply chain and domestic requirement)
• CNA — Drone screening saves 3 to 5 times the effort on high-speed rail bridge inspection (in Chinese; 252 of 350 km on structure, biennial inspection, 700 m/day manual, 20–25 minute sorties, 4 teams of 9 total)

Power infrastructure

• CNA — Taipower deploys drones to inspect transmission towers (in Chinese; ~20,000 towers and 13,000 km of line, how infrared temperature measurement changed, the 2017 Hoping tower collapse and 580,000 affected households)
• CNA — Taipower's drone management system enters service (in Chinese; the 65 → 12 US-made → 44 Taiwan-made fleet transition under cybersecurity policy, the management system's compliance alerting)
• CNA — No more Spider-Man on the towers: Chia Hui Power launches fully automated drone inspection (in Chinese; first fully automated inspection plant, 75% headcount and 50% cost reduction, no outage required, maintenance history database, 5G private network)
• CNA — Hung Kuang University inspects high-mountain towers by drone (in Chinese; AXH-E230 across the Central Mountain Range, 46.5 km and 79 towers in one hour, "silent flight" autonomous navigation with the link down, hydrogen fuel cell medical resupply)

On this site

• The Drone Industry Map: Components, Regulatory Ceilings, and the Non-Chinese Supply Chain Rebuild
• BVLOS in Three Jurisdictions: Taiwan Has No Framework At All
• After "NT$2M a Year Flying Drones": Agricultural Spraying Has Run a Full Cycle
• Search and Rescue Drones: The One Application Whose ROI Isn't Money — and the Easiest Budget to Cut
• Taking Apart Two TTSB Crash Reports: Neither Was the Operator's Fault
• Four Gates into Taiwan's Drone Industry: The Entry Mechanics Public Records Can Tell You