CS-425 is the family's crawler for passages that already exist: recon, mapping and locating in tunnels and culverts, on a fibre with a person on the far end. It can interpose. It never strikes.
The Underground family divides in two. CS-410 and Aquifer CS-411 make the hole; CS-425 works holes and passages that already exist.
There is no bore head on this machine and no cutting face. It enters tunnels, culverts, storm sewers and service crawlspaces, and its job in them is recon, mapping and locating in confined dark spaces that no satellite fix reaches and no radio crosses.
The battery is aboard, and the fibre carries comms and guidance rather than power. That is the split against the boring line, where the tube feeds the head.
The boring line. It drives a bore where there is no route, and leaves the tubing in the ground behind it as conduit.
The passage workers. They enter, listen in, crawl or ride ground that is already open, and none of them cuts to get there.
The family's one physical response: the machine's own mass, set down where a person has confirmed it should be.
A tunnel robot's useful cross-section is set by the worst passage on the mission, not the average: partial blockages, offset joints, root intrusion, standing water.
Three geometry rules come out of that, and each is a requirement rather than a feature. The stance is low and wide inside a circular envelope, the track base as wide as the invert allows, because a curved floor is a tipping problem. The running gear is top-bottom symmetric, because a crawler that cannot run inverted after a rollover in a pipe nobody can reach is a lost unit. And nothing presents a rearward-facing shoulder that can catch a joint lip, because reverse-out along the laid fibre is the primary recovery mode.
| Item | CS-425 |
|---|---|
| Chassis footprint | ~300 mm W × 260 mm H × 520 mm L |
| Mass | ~25 kg |
| Passage clearance | Tunnels and culverts, ~450 mm diameter and above |
| Crawl speed | ~0.8 m/s · ~2.9 km/h |
| Battery endurance | ~4 h continuous recon |
| Drive | Tracked, dual electric drive |
| Tether | ~500 m of fibre per spool, breadcrumb return |
| Sensors | Forward and aft cameras · lidar and mapping (SLAM) · gas sensor (O₂, CH₄, CO, H₂S) |
| Guidance tiers | Fibre-tethered teleoperation, the default · mapping (SLAM) |
| Autonomy ceiling | Mapping autonomy only. Engagement autonomy does not exist |
| Emissions, any tier | None. No radio, no GNSS |
| Payload | Sensor suite only. No warhead, no fuze, no energetic material |
A crawl speed is not a transit rate through real passage; the interpose clock below is worked at the effective rate.
The class it is sized for is the culvert, utility-tunnel and sewer class, where a mature commercial inspection-crawler industry has already proved the packaging closes. That is what a real network offers under a facility, a road or a perimeter.
The structural problem here is the opposite of an interceptor's. No launch shock and no aero loads, and instead continuous immersion in ground that is wet, abrasive, corrosive, dark and actively trying to jam the mechanism with silt.
The governing load cases are a rollover, a snag during reverse-out and a season of abrasive silt in the seals. None is a stress case. The floor is temporary immersion at IP67 class rather than splash-proofing, because a dry tunnel still means standing water, spray and saturated silt. The hard sealing problems are the rotating ones: the track drive output shafts and the point where the fibre leaves the body. Sewer-crawler practice sets the precedent, pressurized housings with monitored internal overpressure, which turns a seal failure into a live abort criterion.
Chassis, track frames and any load-bearing shell are moulded from the fleet's structural resin on the group's own five-axis mould shop. Covers, spool flanges and handling fixtures are rPET, scoped to the parts that carry no drive loads.
Corrosion-immune in sewer atmospheres, hydrogen sulphide and chlorides, where steel inspection crawlers fight rust. Metallics stay where polymer cannot go: axles, gears, fasteners, seal interfaces.
Thermal design is the other consequence of a sealed box. Dead tunnel air gives no forced convection, so every watt of the electronics load leaves through the shell by convection, conduction through the tracks and radiation, at roughly 15–20 K of shell rise at the working power class. Mud coating the shell insulates it, so the answers are the chassis as heatsink and duty-cycling the lidar, which is what the energy budget asks for anyway.
Twin tracks, skid-steered, each independently driven. It is the configuration the whole sewer and pipe-inspection industry settled on for floors that are wet, silted and stepped.
Obstacle capability scales with sprocket and track height: the climbable step is roughly the same order as the front sprocket radius, which is a statement about the machine rather than about the ground. The ground is the harder half. Tunnel inverts are the worst traction surface in the fleet: wet biofilm on concrete, loose silt over hardpan, sloped culvert walls.
| Surface class | Traction coefficient | What it costs |
|---|---|---|
| Firm invert, thin film | ~0.10 | The easy case. The drivetrain is barely loaded |
| Silted or muddy floor | ~0.25 | Two and a half times the load, and the surface where a track run packs |
Terrain classes. Drivetrain sizing is easy at this scale: traction, not torque, is the problem.
Three consequences follow and the design carries all three. Slip on biofilm is large and asymmetric, so the locating stack below never lets track odometry carry the budget alone. Grade is stated on wet biofilm rather than on a dry floor, because culverts and laterals run at real slopes. And silt packing in the track run is the classic inspection-crawler failure, so open self-clearing geometry beats track guards sealed for neatness.
On a small crawler the sensors and the computer out-consume the wheels, and that single finding sets the way the machine is worked from end to end.
A crawler, unlike anything airborne, pays nothing to stop: locomotion power on station is zero and the brakes are passive. The electronics load runs whether the machine moves or not, and lighting is a first-class consumer of it, because a tunnel is absolutely dark and nothing is seen that is not lit.
| Continuous load | Draw |
|---|---|
| Lidar | ~5–10 W |
| Compute and SLAM | ~5–15 W |
| EO/IR | ~2–4 W |
| Illumination | ~2–10 W |
| Telemetry, battery management and the rest | ~2 W |
| Continuous total | ~16–40 W |
Component classes. Endurance is bought by duty-cycling the lidar and the lights, not by drivetrain efficiency.
So the doctrine that falls out of the budget is creep, stop, look: move slowly, halt often, sense hard while halted with the lidar duty-cycled. Speed costs almost nothing in energy and a great deal in traction risk, payout tension and map quality. Endurance is ~4 h of continuous recon, and 30% of the pack is held for egress, which is a range rule before it is a battery rule.
Two architectures exist for a tethered crawler, and choosing between them is what decides how far one can go.
Surface-spooled, the console pays out and the robot drags the whole laid length behind it. A dragged tether's friction grows with that length and every corner multiplies it, the capstan effect, which is what actually range-limits a dragged-tether robot. CS-425 is robot-spooled: the spool is aboard, the fibre is paid out stationary onto the floor behind it, and it lies where it fell. Drag does not grow with range and corners do not compound it. What it lays is the breadcrumb back out, ~500 m per spool.
There is a cost. The spool rides the vehicle, the fibre is committed as it goes, and a snapped fibre ends comms, so egress then runs on the machine homing on its own map. That is what sets the minimum onboard autonomy even on a platform whose doctrine is a person driving.
| Fibre class | Diameter | Mass per km | Wound volume per km | Read |
|---|---|---|---|---|
| Bare or acrylate | ~0.25 mm | ~0.1 kg | ~0.06 L | Trivial to carry, fragile in mud and abrasion |
| Tight-buffered | ~0.9 mm | ~0.7 kg | ~0.8 L | The working class: it fits on mass and on volume |
| Ruggedized mini-tether | ~2–3 mm | ~4–8 kg | ~4–9 L | An envelope breaker at a kilometre |
The spool closes at buffered gauge, so fibre survivability on abrasive wet floors is bought by payout placement and routing rather than by jacket armour.
One fibre carries the teleoperation video, the lidar stream, the telemetry and the commands together. Bandwidth is effectively unlimited against this sensor suite and latency is nil at these lengths.
Underground the hard problem is not driving. It is knowing where the machine is, answered with no external reference of any kind.
An inertial unit, track odometry and paid-out fibre length. Odometry slips and a MEMS inertial unit drifts, but the spool adds a third and unusual input: paid-out fibre length is a path-length odometer that does not slip. Fusing the three bounds along-path error better than any alone.
The mapping sensor doubles as the locating sensor. Long uniform pipe is the pathological case for scan-matching: self-similar geometry gives perceptual aliasing and along-axis ambiguity, and that is where the fibre-length odometer carries it. Junctions, joints and defects are the landmarks.
The tether's own paid-out length, counted at the reel. It is the group's own locating practice, thirty years of steering tools through ground nobody can see, and it ties the map to the portal the crawler went in at.
A denied surface loses the third layer and the first two carry the budget alone, which is what makes the fibre odometer matter more than it looks: it is the one input a slipping track and a hundred metres of identical pipe cannot both fool.
The move is the fleet's, and the boring line's tube-to-head seam is the nearest example: define one payload bay once, so that a new variant is a packaging change rather than a new vehicle.
The bay is a mechanical envelope, a power budget, a data interface and a mass cap, settled once. The core set is always fitted: an EO/IR camera with active illumination, a thermal imager and a small lidar. The thermal channel earns its place here, because ground like this is a favourable infrared environment, where a warm target or a recent heat trace stands against cold walls.
The lidar record fused with the locating stack is the product: tunnel-network geometry and condition assessment. The civil twin is this variant with reporting software, so the defence build and the civil one are the same build.
Oxygen, carbon monoxide, hydrogen sulphide and methane at the lower explosive limit: is the passage survivable, and has there been recent combustion or equipment activity. It is also the civil confined-space-entry survey product.
Carries and releases CS-430-class listening nodes at chosen points, extending a Trembler Line into passages no surface emplacement reaches.
A parked unit whose spool end becomes a network drop, so later units and emplaced nodes tap the laid fibre as backhaul.
Every module on the list is a sensor, an annunciator, an emplaced node or an inspection package. None of them carries energetic content, for any customer, and any proposal beyond light, sound and obstruction is a counsel-first corporate question rather than a product decision.
CS-425 is the one platform in the family designed to touch an adversary at all, and the way it does that is by parking its own mass in a passage too narrow to go around.
Its only effect is its own mass. Nothing detonates, nothing fires, nothing is thrown. It carries no warhead, no fuze and no energetic material, in any variant, and that boundary binds the whole fleet. The interpose fit adds high-intensity light and sound to the obstruction: presence, warning and denial by being unmistakably there, non-injurious by design, and switched on by a human command over the fibre.
Autonomy stops at the map, deliberately. The machine builds its own passage map and creeps a known run on SLAM, and it cannot decide, on its own, to move into a block position. When a cue arrives over the fibre it presents that position as an option. A person confirms before it moves in and holds, and a person confirms again before any physical engagement follows. There is no autonomous engagement tier in the design to enable, gate or unlock.
No radio and no GPS anywhere in the design. No receiver to jam, no fix to spoof. The one physical response the family owns is a machine parked in the way, on a person's confirmation, in a passage nobody had to walk.
The interpose role depends on the Trembler Line CS-435, a buried picket of strung CS-430 nodes. It detects, it localizes, it cues. It never strikes.
Detect, localize, cue, interpose ahead of the path. A person stands between the cue and the movement, and the machine waits out on the network rather than at the portal. Through real passage, at the creep-and-halt duty the energy budget rewards, the effective transit rate is about ~0.2 m/s, and the clock reads from there.
| Standby-to-interpose distance | Transit time | Read |
|---|---|---|
| 100 m | ~8 min | Plausible against slow threats: digging, walking |
| 250 m | ~21 min | Needs early picket detection |
| 500 m | ~42 min | Only works with long-range early warning |
At an effective ~0.2 m/s through real passage. The design consequence is that it does not sprint to intercept.
Pre-positioning and picket density buy the interpose timeline; vehicle speed does not. Roughly one standby unit per few hundred metres of defended network is the shape of it, so the sensing line and the response element are bought as one system. The nodes are on CS-430 and the line they make on Trembler Line CS-435.
There is no recovering a dead crawler 400 m into a ~450 mm pipe. So the recovery system is the machine itself, run backwards along the fibre it laid on the way in.
The safety chain matches the rest of the fleet: a hardwired e-stop, a wired stop over the fibre, and a loss-of-link rule written for confined space that ends in stop, hold, wait, silent. It never pushes forward blind through unmapped passage. If a leg ends early it holds exactly where it stopped, silent, waiting for a person, and recovery is a walk down the fibre rather than a search.
It suits an owner of ground with passages under it, and a perimeter a standing post covers today. It does not answer a weapon requirement.
If you hold ground with a network under it, the conversation starts with the passage rather than the machine.
A screened enquiry is the way in, and engineering answers it, not sales. Bring the network you have to know: the passage classes, the access points, how far apart they sit and what is at the far end. We bring the crawler, the fibre and the arithmetic that says where a standby unit sits for the clock to close. Counsel comes first on any international transfer and permits are taken per shipment. Every design in the family is patent pending.
The Underground family is the whole line side by side. Trembler Line CS-435 is the picket that cues this machine and CS-430 is the node it is strung from. Who it is for is the buyer's view.
All Underground family designs are patent pending; international transfer is export-controlled and subject to Canadian government permits taken per shipment, and nothing on this page is an offer.
