CONFINED INFRASTRUCTURE ROBOTICS · HONG KONG × SHENZHEN
PipeGuardian combines a collision-tolerant inspection robot, topside link and recovery, spatial sensing, and a segment-level evidence platform for closed pipelines and confined infrastructure — shrinking the unknown before people enter.
Field application exists · concept / lab / field stay distinct · no invented performance, customers or certificates
THE PROBLEM
For closed networks, utility corridors, drainage tunnels and industrial passages, the costly moment is not “how we repair after we see it”. It is the inability to prove, before entry, whether the atmosphere is tenable, the path is recoverable, and the record can be audited.
Hong Kong and Shenzhen already have large underground inventories. Many high-risk segments still depend on limited human entry, CCTV crawlers or temporary lighting. Crawlers are mature on long, wet, standardised runs. Once a job needs a shaft change, local collapse, debris, weak structure or a fast three-dimensional context, deploy time and human exposure rise. An open-air drone cannot answer an in-pipe question: GNSS fails, the link fades, contact is often unavoidable, and the recovery path has to be designed first.
Industry already has comparable paths. Tethered confined-space drones such as ScoutDI-class platforms bind power, data and recovery to a cable, and suit long observation in tanks, sewers and tunnels. Hong Kong EMSD public materials on confined-space robots and PipeScan AI show local utilities writing robots, imagery and defect assist into confined-space work. Groups such as PKU Nanchang keep publishing mechanisms and sensing for gas and drainage pipe robots. Those facts prove demand is real. They are not PipeGuardian field results.
PipeGuardian’s judgement: the Greater Bay Area is short of a mission system that binds pre-entry reconnaissance, loss-of-link handling, gas/thermal/image evidence and asset coding onto the same segment — not “another machine that can fly”. Flight is one way to take evidence. If the vehicle cannot be recovered, the weak link cannot be explained, and a qualified person cannot review the clip, the flight has no product value.
SYSTEM, NOT A GADGET
The product is a closed loop of safe work, robot, data and decision. Flight is one way to collect evidence. Recovery, communications, human takeover and the asset interface belong to the product as well.
Low-light / night vision for dim imagery, thermal for heat anomalies, dedicated methane / CO / H₂S / VOC sensors for gas evidence. First edition enters controlled, de-energised, non-explosive scenes only.
Start with tethered or wired backup and a small number of recoverable nodes. Record weak-link and loss-of-link locations. No promise of a full underground mesh.
Power, data offload, health checks and manual recovery. In-pipe battery swap is a later milestone.
Synchronise video, point cloud, gas and pose. Keep model version, confidence, human review and the raw files.
Plans a quick-release fire-suppressant and insulation-tape module. Today only the interface and a safety gate are defined. Independent tests, certification and scene approval come before a controlled trial.
INSIDE THE INFRASTRUCTURE
PipeGuardian is built for large pipe tunnels, underground water networks, drainage, utility corridors and industrial passages from the inside — not façades or open-air patrol UAVs.
Pre-entry reconnaissance, crack / foreign-object / heat-anomaly marks, and recoverable inspection. The aim is not to replace all civil testing. It is an auditable space and atmosphere snapshot before people go down.
Validate first on isolated, dry, non-explosive segments with a clear entry and exit. Wet, sewage and corrosive environments need separate seal, decontamination and materials gates. They are outside the first promise.
Bind video, depth, gas and pose to a segment and an asset code, so a crew can turn a “suspect clip” into a work order instead of a pile of unsearchable video files.
Full scene notes sit on applications. Drainage / corridor / industrial envelopes sit on the use-case pages.
METHOD
Every performance figure starts as a target. Repeated tests, raw logs, failure records and customer acceptance turn it into a fact. A number without a date, conditions, a reproduction method and an owner does not appear in public materials.
G0—G2
Confirm the scene is real and the exclusion list is clear. Then prove the cage can take a hit, be repaired, and spare the rotors. Then prove, on a 30 m straight and a bend, that loss of link aborts and position can be replayed. Until those three gates pass, there is no endurance advertising.
G3—G4
Video, depth, gas and pose must align to a segment. Defect-assist models only rank candidates. They keep a held-out set, miss cases and human review. Thermal does not replace gas. Gas does not replace a structural judgement.
G5—G6
G5 asks, on paired tasks, whether coverage, time, human exposure and decisions actually improve. G6 is reliability, FMEA, maintenance and certification files. The 18-month plan aims at G5. G6 is defined with the actual scene and the certification body.
SAFETY ENVELOPE
The first verifiable envelope excludes pressurised gas, a fully flooded pipe, explosive mixtures, a critical line that cannot be recovered, and active intervention. PG-X is a research interface, not an approved field-response capability.
FIRST BEACHHEAD
Prefer isolated drainage, utility corridors and non-explosive industrial-park passages. Hold pressurised gas, flooded pipe, explosive mixtures and unrecoverable production lines. That is not market avoidance. It puts the first auditable evidence where a test can be repeated.
MVP mission envelopeHONG KONG × SHENZHEN
Use Hong Kong universities, science parks and public-sector trial mechanisms for credible evidence. Finish mechanical prototypes, supply chain, small batch and Greater Bay Area delivery in Shenzhen. Costs, people and IP must split, or subsidies cannot be stacked in discussion.
HK–SZ operating model18-MONTH MISSION
The planned cycle serves internal governance. It is not a delivery promise to customers or funding agencies. Any public date binds to an evidence-gate status.
Modular cage, well-head station, controlled straight and bend flight. The failure list matters more than a success clip.
Video, depth, gas and pose in sync. Defect marks trace to device, time and model version.
Paired process for coverage, time, human exposure and decision value. A demo reel does not replace acceptance.
COMPARE HONESTLY
Crawlers, tethered drones and collision-tolerant aerial vehicles do not cancel each other. PipeGuardian takes the third as the in-pipe reconnaissance main path, and treats the first two as industry baselines that must cooperate.
INSIGHTS
Until a trial is accepted, this site does not publish success cases or a customer list. Insight articles explain method, bounds and public sources, so reviewers, collaborators and engineers can see what we are studying.
Object, window, evidence index and review deadline as a file the next shift can pick up.
Read insight → CADENCEWrite inspection as an asset calendar: pre-entry, capture, triage, review.
Read insight → INDUSTRY MAPFrom tanks and sewers to corridors and industrial passages, permit, recovery and evidence chain decide the form.
Read insight → METHODHow G0–G6 constrain the website, the pitch, subsidies and media so the materials do not fight each other.
Read insight → HK CONTEXTA public-sector trial wants an acceptable task, not a maxima poster.
Read insight → LOCALISATIONA degradation alarm sits closer to real engineering than a centimetre slogan.
Read insight →