The hard part of a pump-out is the yard.

It’s 8:10 on a Wednesday morning in late April, and a vacuum truck is idling at the curb of a split-level outside Lansing. The work order says the house has a 1,000-gallon tank, last pumped in 2021. It doesn’t say where.

The driver walks the yard with a steel probe rod, pushing it into the lawn every few feet along the line where he guesses the sewer leaves the house. He hits concrete on the third try along the back fence. Four of the six screws on the riser lid come out cleanly, and the fifth is rusted solid. He drags 100 feet of 3-inch hose from the truck around the garage, kneels at the open riser and breathes the gas coming off the tank while he gets the hose in.

The pumping is the short part. The truck does it. Almost everything else on that job is one person working alone in somebody’s backyard.

Where a crew’s time goes

Anyone who runs a pumping company knows the list. First visits rarely come with drawings, so the tank has to be found by probing and digging. Lids are screwed down, cracked, buried or seized. The hose is heavy and has to go out and come back on every job. And the riser is where the hydrogen sulfide is.

That last one matters most. According to NIOSH, hydrogen sulfide deadens the sense of smell above about 100 parts per million, so the rotten-egg warning disappears just as the gas gets more dangerous, and concentrations above 1,000 can knock a person unconscious. An analysis of OSHA fatality records from 1984 to 1994 counted 80 deaths from hydrogen sulfide. Nineteen of them were coworkers trying to rescue someone.

The people doing this work are also hard to hire. Septic driver postings typically ask for a Class A commercial license, or a Class B with a tanker endorsement, and in 2025 and 2026 they’ve offered roughly $23 to $28 an hour.

A market that comes back on a schedule

The EPA says about one in five American households relies on a septic or other on-site system, and that a typical household tank should be pumped every three to five years. Fixr put the national average for a routine pump-out at about $400 in 2025, with most jobs between $300 and $600.

That works out, by my estimate, to somewhere between 5 million and 9 million residential pump-outs a year, or $2 billion to $3.5 billion in pumping alone, before inspections and repairs. The work is local and recurring, and private equity has noticed. In August 2025, Gryphon Investors’ Wind River Environmental added M&S Septic and Fenkner Septic in Pennsylvania to a growing roll-up. Owners who buy routes want more jobs out of each truck and each driver.

What I’d build

I’ve designed a robot for the yard part of the job. It’s called SCARAB: a tracked rover about the size of a riding mower, with an industrial arm, a set of swappable tools, ground-penetrating radar and four cameras, plus a dock that rides on the truck’s hitch and carries 115 feet of 3-inch hose on a powered reel.

The truck operator still drives the route and still parks at the curb. On site, the job runs like this.

1. 01 · FindNo drawings neededIt maps the yard as it drives, spots riser lids, reads their lettering and confirms the buried concrete with radar.
2. 02 · CoupleThe truck you ownIts arm couples the hose to the truck’s standard 3-inch camlock inlet, the same one a person uses.
3. 03 · OpenLid off, gas checkedIt lays hose to the riser, backs out the lid screws, works seized ones loose and reads the gas before anyone gets close.
4. 04 · PumpThen put it all backIt pumps while tracking the tank level, closes and torques the lid, rewinds the hose and sends a report with photos.

The operator taps a tablet three times: to confirm the yard is clear, to start the vacuum and to stop it. Nobody kneels at the riser. Each tank it finds is saved, so the next visit, three or four years later, starts at the lid.

What it did in simulation

SCARAB exists today as a full engineering design and a physics simulation, and I want to be precise about which is which. The design is 48 custom parts in CAD with drawings, a 141-line priced bill of materials, and calculations for stability, traction, hydraulics and energy. The simulation drops the robot into a yard it has never seen, with a buried tank, a seized lid screw and a work order, and lets it do the whole job.

From the October 2026 simulated run. The fourth of six lid screws was seized on purpose; the robot hit its torque limit, rocked the screw back and forth and freed it.

In that run it found the tank without help. It mapped the yard, spotted two green riser lids from across the lawn, read “SEPTIC” on them so they wouldn’t be mistaken for a water meter, and confirmed concrete 32 centimeters down with its radar. It worked out where the drainfield was from the outlet side of the tank and kept its tracks off it. At the riser, its gas monitor read 18 parts per million of hydrogen sulfide and sent a warning to the operator’s tablet before pumping began.

What it has to earn

The bill of materials puts the first unit at about $236,000. At 10 units the cost falls to about $197,000, and at 100 to about $161,000, from volume pricing on parts and lower assembly cost as builds repeat. Running costs are small: about $76 a month for cellular data and cloud services, and roughly half a kilowatt-hour of electricity per job.

My arithmetic, not an industry figure: straight-line over five years, before financing and service.

Seven more jobs a month per truck is the bar, or about one more every three working days. For comparison, a helper at $23 to $28 an hour costs $4,000 to $4,900 a month before taxes and benefits. A company could use the robot either way: to add jobs to a one-person truck, or to send one person where it used to send two.

The objections

Yards are a mess. They are, and a simulation doesn’t settle it. Wet clay weakens ground radar, and slopes over 10 degrees with the full hose out need testing. Both are on the pilot’s list of things to prove, along with bringing the rover’s weight down from 440 kilograms toward a 300-kilogram target.

Every lid is different. The pilot starts with screw-down riser lids, the kind most newer systems have, then side-screw and concrete lids. Tanks with buried lids that need digging stay with the crew for now.

The truck is already full. The rover and dock together weigh about 658 kilograms, which would cost roughly 169 gallons of payload if carried on the truck. For the pilot they ride on a 6-by-10 tilt trailer. A frame-mounted carrier can come later as a factory upfit for new trucks.

Gas and electronics don’t mix. The electrics that work over an open riser are specified to hazardous-area ratings. Two safety laser scanners watch for people and pets, and four emergency stops plus a wireless stop feed a safety controller that runs independently of the main computer. Hazardous-area certification and UL and EMC testing are part of the budget.

Rural cell coverage is bad. Driving, arm control and every stop rule run on the robot itself. If the signal drops, it finishes whatever is safe to finish and hands the next decision to the operator.

How I’d find out if I’m wrong

I’d prove it in stages, and each one has a number that can end it.

• First, one unit on a test bed. Build it from the bill of materials and run the riskiest parts until they’re boring: the camlock coupling must connect on at least 49 of 50 tries, the lid tools must open the five most common lid types, and the hose must pay out and rewind 100 times without a fault.
• Then 20 real tanks with one partner company. Over four to six weeks, the robot rides along and works the job while the crew stays responsible for it. It passes if it finds at least 17 of the 20 tanks on its own, opens every lid type it’s rated for at least 80 percent of the time, finishes within 25 percent of the crew’s time, and has no safety incidents.
• Then three units on routes for a season, against matched trucks without one. The one number that matters is completed pump-outs per truck per month. If the robot trucks don’t add at least seven a month over their matches, or if they cost more crew hours per job, we stop.

The matched trucks matter because a partner will naturally give its best driver the new machine. Comparing the same routes, the same season and similar drivers is the only fair way to see whether the robot is doing the work.

If it works, the next step is ten units with that partner, priority on the first production run, and a design shaped by the lids, hoses, trucks and yards its crews actually see. If it doesn’t, the partner has spent a season and gained a map of every tank on its routes.

Back at the split-level outside Lansing, the driver would still be the one at the curb. He’d tap the tablet to confirm the yard was clear, tap it again to start the vacuum, and tap it once more to stop it, and he’d never have to kneel at the riser.