The idea

Picture an autonomous robot that walks the Australian bush with a metal detector, marks every spot that might be gold, and does it all night without getting tired. That is Happy's idea. I am treating it as a proposal, and this post checks how far it is from real.

What exists today

Detectors already guess for you. One new gold detector shows a Gold Chance Indicator that leans toward probable gold or toward iron junk.[9] A person still has to dig. Robots with sensors also exist. A team at the University of Central Lancashire built an autonomous drone that carries a magnetometer to find buried landmines.[7] A magnetometer senses magnetic metals, and one maker says plainly that it cannot detect gold nuggets.[6] So the drone is the wrong sensor for this job. A 2018 engineering paper looked at prospecting robots built on inductive metal detectors and called them a possibility.[8] I found no finished product that does this.

What the rules assume

In Western Australia, metal detecting counts as prospecting, and each person needs their own Miner's Right.[2][3] That permit allows prospecting on Crown land with hand held tools such as a metal detector, and keeping samples up to 20 kilograms.[1] It is not a mining tenement and does not authorise heavy machinery.[2] You must also fill any holes you dig.[1] Victoria also requires a miner's right, and says mechanical equipment is not permitted.[4][5]

Rules differ by state, and I checked only those two. I found nothing that says whether a robot counts as a hand held tool. The permit also belongs to a person, so who holds it when a robot does the walking? That is an open question. A real find would also lead somewhere else. A permit for samples is not a license to mine.

The hard parts

Finding a signal is only the start. A robot must tell gold from the iron scrap that old goldfields are full of, dig a hole, check it, and fill it back in. Legs, wheels and a small digging arm are harder than the detector. That is why this is a proposal, not a product.

If the robots share finds, they have something worth hiding. How they could keep targets private on a license free radio band is covered in Radio for Signed Payments. A machine that earns its own money is the subject of When AI Has a Wallet. A machine someone else can take over is the subject of Who Reflashed the Robot?

What it costs and what to expect

Gold was about $3,861 an ounce, or $124 a gram, in mid November 2026.[10] A Boston Dynamics Spot starts at about $75,000.[11] A cheaper robot dog, the Unitree Go2 Enterprise, costs $8,900.[12] A top gold detector, the Minelab GPZ 7000, was $9,499 on sale at one US shop.[13] A cheap robot plus detector is about $18,400, which is about 150 grams of gold. A Spot plus detector is about $84,500, which is about 680 grams.

What a detector finds is another matter. Most finds are tiny flakes or pea sized pieces that barely cover fuel for a day.[14] Big nuggets are rare. Victoria saw a 4.6 kg one in 2023 and a 119 gram one, worth about $14,800 at today's price.[4][18] I found no reliable average, so these payback times are my assumptions, not data. At one gram of gold a night, the cheap setup pays back in about 150 nights. At 0.2 grams, about two years. At a flake of 0.05 grams, about eight years, before any repairs.

The Mars rover idea

Rare finds matter less to a patient machine. NASA's Opportunity rover was designed for 90 Martian days and instead lasted 15 years and drove more than 45 kilometers on solar power.[15] Its twin Spirit ended when it got stuck in sand.[16] A solar prospecting robot could also work for years, but it needs a rescue plan. A person with a truck can drive out and fix a stuck robot, which NASA could not do.

Speed is the catch. Assume a robot moves 0.1 meters a second with a coil about 0.36 meters wide, running 8 solar hours a day. It sweeps about 31 hectares a year, or about 3 square kilometers in 10 years. The Golden Triangle covers about 9,600 square kilometers.[17] A fleet of 100 robots would cover about 3 percent in 10 years. Detectors also reach only about a meter down.[19] These are my assumptions, not measured figures. The number nobody has given me is how many nuggets lie per hectare.

Sharing paths, swapping batteries, running at night

Happy's idea is that robots report their paths so no two search the same ground. At this scale accidental overlap is rare, so the map matters most when many robots chase the same promising patch, or when one reports it is stuck. It also makes a second pass with a deeper coil a choice, not an accident.

Battery swaps help too. A depot with solar panels and spare batteries can charge by day, and a robot swaps in a full one at dusk. A support truck can do the same and also rescue stuck robots. Running 24 hours a day triples the ground covered, to about 93 hectares a year. Each extra hour a day adds about 4 hectares a year, so a battery that lasts only part of the night still helps. But moving three times as far wears wheels and motors faster. Opportunity's wheels made more than 90 million turns over 45 kilometers while moving very little.[20]

The most practical version has robots sweep and flag spots while a person with a truck digs, swaps batteries and fixes problems. A Miner's Right belongs to a person, so that is clearer than a robot digging alone, though I found no source saying a regulator would agree. Health reports could use a protocol such as SNMP, the Simple Network Management Protocol. Over a slow radio link a lighter one may fit better, and I have not tested either.

Keeping it safe

Kangaroos are unlikely to hurt a robot, and a thermal camera that makes it slow or stop protects the animals. Theft and vandalism of an unattended machine worry me more. A tracker, a tamper alarm, a siren, a camera that sends photos to a phone and a body that is hard to carry off are the defense, and any decision to act belongs to a person. The map of finds should be encrypted, as in Radio for Signed Payments.

Where to search and how to hold the ground

Walking free ground with a Miner's Right is one thing. Holding ground is another. In Western Australia a prospecting licence covers under 200 hectares, must be marked out, and lets the holder extract or disturb up to 500 tonnes of material.[21] An exploration licence can cover a much larger area, up to 21,700 hectares, made of set blocks.[22] A fleet that sweeps thousands of hectares a year would need ground like that, or a lot of permission from whoever already holds it.

In the United States, the Bureau of Land Management treats casual use as activity with negligible disturbance, such as collecting specimens with hand tools or hand panning, with no mechanized equipment or explosives.[24] The federal rule says casual use generally includes metal detectors and other battery operated devices for sensing minerals.[23] A robot that walks, senses and flags a spot sounds close to that. A robot that digs is where it gets unclear, and I found no source that says whether a robot counts as hand tools. I did not check state land, national forests or how claims work on top of this.

Ways to dig, with figures

Happy says the robot is only the detector. It sweeps, logs a GPS location for each target, and people dig those spots by hand. That is the version I would build first. He also pointed out that the robot could carry a digging arm, or that a separate purpose built digging robot could do the digging. Here are the three side by side, kept simple.

VersionThe robotWho digsCost and effortRules
Sense and flagSmall motorized detector, no armA person with a hand tool, once per flagCheaper robot, but a person walks out to every flagClosest to a detector, and hand digging fits the permit wording
Sense, dig, check and fillDetector plus a digging armThe robotDearer, with more parts to break, but less walkingDigging is where the machinery wording bites, and it is unclear
Sense and flag, shared diggerSmall motorized detector, no armA separate purpose built digging robot that visits each flagOne digger costs more but serves many detector robotsSame machinery question as the arm

Here are rough figures for the rational case, 31 hectares per robot a year. Every number is my guess. Assume 10 flags per hectare, most of them iron junk, 10 minutes per flag to walk, dig, check and fill, and $50 an hour for a person, all in. That is about 310 flags and about 52 hours of digging per robot per year.

OptionExtra equipmentDigging cost per detector robot per yearNotes
Person digs by handNoneabout $2,600One person digging about 1,500 hours a year could serve about 29 robots, close to the 33 in the table above
Arm on each detector robotabout $10,000 eachabout $1,000 a year upkeep, no digging laborHeavier, more to break, and the machinery question
One shared digging robot for about 20 detector robotsabout $50,000, or $2,500 per detector robotabout $500 a year upkeep shareSimple, but it must drive to every flag

Spread over five years, the arm costs about $3,000 a year all in, which is $1,000 of upkeep plus $2,000 of purchase, against about $2,600 for a person. That is roughly a wash. The shared digger comes to about $1,000 a year, a saving of about $1,600 per detector robot, if it works. All of this is guesses, and none of it changes the sign. In the rational case the fleet still loses money unless the ground holds about 1 gram per hectare. If flags run to 30 per hectare, a person's digging bill triples to about $7,800 a year per robot, which is why the junk rate matters.

Victoria lists electronic detectors among the permitted tools and says no machinery is permitted. The page does not define machinery.[5] Western Australia says a Miner's Right does not authorise heavy machinery,[2] and the US rule says casual use involves no mechanized equipment or explosives.[24] Happy's reading is that this means digging and ground moving equipment, such as an excavator, because otherwise nobody could drive a vehicle to the site. That is a sensible reading and it is mine too, but it is a reading and not a ruling. I found no definition, and no regulator has ruled on a robot. A small motorized detector looks closer to a detector or a vehicle than to an excavator. A digging arm is the harder case.

A sensing only robot is also the lowest impact version, because it does not disturb the ground. The Bureau of Land Management says advance contact may be needed where the cumulative effects of casual use could cause more than negligible disturbance,[24] and a fleet using the same tracks every day could add up. That is my reasoning. Vehicle access has its own rules, such as tracks, landholder consent and park limits, and I have not sourced those.

One practical catch is location. The coil is about 0.36 meters wide, and ordinary GPS can be off by several meters, so finding the same spot again would likely need a centimeter grade GPS, often called RTK. I have not sourced that. The bigger unknown is the cost of digging: how long each flag takes and how many flags are iron junk. I have no data on either, and they could matter more than the robot's price.

Depth and the one big nugget

Happy heard that some detectors reach 8 feet. I found no support for that. The deepest figure I found for a top gold detector was nuggets down to about 1 metre, from a retailer blog.[19] My reasoning, not a source: the signal from a small target drops quickly with depth, so a larger coil buys depth but costs speed and power. I also found nothing saying military or non compliant units go deeper, and a detector that breaks radio rules would be its own legal problem. Deeper gold would need a different machine or a person with a bigger one.

Happy's other point is that a fleet only needs one big hit. That is true in principle. Victoria saw a 4.6 kilogram nugget in 2023,[4] which is about $571,000 at the mid November gold price, by my arithmetic.[10] To show the shape of the lottery, suppose one find like that turns up per 500,000 hectares swept. That number is invented. A fleet of 100 robots sweeping about 3,100 hectares a year would wait about 160 years for it. A rich find is possible, but a business plan cannot rest on it.

The honest numbers

Happy asked for honest figures, so here are three cases. Every input is my assumption. The one that matters most, gold per hectare, is a pure guess because I found no data. All money is in US dollars at $124.15 a gram.[10]

ConservativeRationalAggressive
Robot all in cost$85,000$30,000$18,400
Ground swept per year15 hectares31 hectares93 hectares
Gold per hectare (a guess)0.01 g0.1 g1 g
Gold found per year0.15 g3.1 g93 g
Value per yearabout $19about $385about $11,500
Running cost per robot per yearabout $8,700about $3,800about $3,200
Net per yearminus $8,700minus $3,400plus $8,300
Payback on the robotneverneverabout 2.2 years

How I built them. Ground swept comes from the 0.1 meters a second robot above. The aggressive column assumes it runs around the clock, the rational column daytime only, and the conservative column half of that to allow for breakdowns. One person with a truck costs about $100,000 a year, shared by about 12, 33 or 100 robots. Holding the ground costs about $24 a hectare a year, a stand in I took from US claim fees and did not source here. Robot cost is a Spot class machine for the conservative column,[11] my guess in between for the rational one, and the cheap robot plus detector for the aggressive one.[12][13]

What it says. In the rational case a robot needs about 1 gram per hectare swept just to cover its running costs, and about 2.5 grams per hectare to pay itself back in five years. The aggressive case works only if the ground really holds a gram per hectare, and nobody has shown me that it does. The honest verdict is unproven. The unknown is grams per hectare on real ground, and one small pilot on known ground would answer it before anyone built a fleet.

Other places: where gold and diamonds have been found

Happy asked whether gold has been found in Canada, Siberia and Antarctica. Canada and Siberia, yes. Antarctica, only traces.

The Klondike rush began on August 16, 1896, when George Carmack, Dawson Charlie and Skookum Jim found gold on what is now Bonanza Creek.[31] The Yukon has produced about 12 million ounces of placer gold.[32] More than 120 years later, no significant bedrock source has been found.[33] In Siberia, gold was found in the Yenisey basin in 1838 and 39, the Lena fields took over after 1876, and the Kolyma fields were developed under Stalin using gulag labor.[34] The Magadan region still leads Russia in alluvial gold, in continuous permafrost.[35] Minor occurrences of gold have been found on the Antarctic Peninsula,[37] but no mine has ever been built there, and Article 7 of the Madrid Protocol prohibits any mineral activity other than scientific research.[36] Frozen ground and distance would make all three hard for a solar robot, which is my reasoning.

The best story for the surface treasure idea is Namibia. In April 1908 Zacharias Lewala, a railway worker who had earlier worked at the Kimberley diamond mine, found a raw diamond while shoveling sand off the line and took it to his supervisor, August Stauch.[38] Lewala was not rewarded, and Stauch applied for a prospecting licence.[40] At first the stones lay on the surface, and hopefuls crawled across the desert on their bellies on moonlit nights, when the stones were easier to spot.[39] By 1912 the Kolmanskop operations supplied about 10 percent of the world market.[38] Then the state stepped in. In September 1908 the German colonial rulers declared a forbidden zone, the Sperrgebiet,[40] and gave sole mining rights to the Deutsche Diamantengesellschaft.[41] The lesson for a robot fleet is about land. Surface treasure goes to whoever controls the ground, not to whoever finds it.

Other sensors and targets

The costly part of the robot is the body, the solar panel, the battery and the support crew. The sensor is the cheap part and could be swapped. Happy suggested other detectors, so this table lists what each could look for. Most rows are my proposals, and the last column says how well each is sourced.

TargetSensorHow it would be foundHow well sourced
GoldPulse induction coilWalk, flag, a person digsSourced above
Diamonds on the surfaceCamera and ultraviolet lampSpot by eye, then sortGeneral knowledge, not checked
Landfill metalsMetal detector and cameraFlag and recoverMy proposal
Iron ore, diamond bearing pipesMagnetometerMap magnetic spots, not stonesPartly sourced, magnetometers read magnetic metals only[6]
LithiumSpectrometer, soil and rock samplesSample a grid, then a lab testGeneral knowledge, not checked
UraniumGamma ray spectrometerSurvey a grid, tightly regulatedGeneral knowledge, not checked
Gems such as opal and sapphireCameraSpot by eyeGeneral knowledge, not checked

Diamonds are carbon and do not conduct electricity, so a metal detector cannot find them. Lithium does not occur as a nugget at all. It sits inside minerals, so a robot would collect samples for a lab, which makes it a survey business and not a find and keep one. Rules also change with the target. A Miner's Right covers gold prospecting, and the Namibia story shows how fast a state can close the ground for something else. This paragraph is general knowledge, not sourced here.

Underwater and power

Happy asked about the seabed, starting with the sea west of New Zealand. Iron sands lie off the North Island's west coast, and vents in the Kermadec Arc and Colville Ridge hold gold, silver, copper, lead and zinc.[27] Polymetallic nodules form in abundance south of New Zealand.[28] On the Chatham Rise to the east, phosphate nodules lie along 400 kilometres at depths of 400 metres or less, an estimated 100 million tonnes, but current economics make mining risky.[29] Mining there is a consent fight first. As of that December 2023 report, the marine consents miners need had been refused,[27] and a phosphate application was declined in 2015.[30]

Nodules are the closest thing to a robot picking items off the seabed. They grow extremely slowly, by up to about 10 millimetres per million years,[25] so they are a one time stock and not a crop. The largest known fields are in the Clarion Clipperton Zone, 4.5 million square kilometres between Hawaii and Mexico, far from New Zealand.[26] Their ecology is contested. A research expedition set out in October 2026 to see how seafloor animals fared more than five years after a deep sea mining test.[26]

Power is the real problem underwater. Here are options from most to least realistic. All of this is my reasoning and general knowledge, not sourced. First, a seafloor dock charged by a solar buoy on the surface, the same swap idea as the truck on land. Second, a tether to a ship or buoy, which I believe real nodule collectors use. Third, a battery that uses seawater as the electrolyte with a metal anode such as magnesium or aluminum. It works, but it burns the anode, so it is a refuel and not a free source. Fourth, algae or mud fuel cells, which are real science but give tiny power, enough for a beacon and not a crawler. I left out plutonium style generators on purpose. They are tightly regulated, the fuel is scarce, and a machine that can be lost or stolen is the wrong place for one.

Where this stops

These are rules as of November 18, 2026, and rules can change. This is not legal advice, and I am not a lawyer. Happy also suggested a mounted firearm, a flame effect, a microwave crowd control device and mace or bear spray to scare off intruders. This post does not cover armed, fire based, chemical or directed energy defenses. A machine that fires or sprays on its own cannot tell a thief from a harmless person, and fire in dry bush is a bushfire risk. I found no law that mentions robots, and I did not check every state or any other country. Check the rules where you plan to work before you send a machine onto any land, and get the owner's consent. Underwater power, diamonds, lithium, uranium and the other targets are general knowledge or proposals, flagged where they appear, and I checked seabed news only for New Zealand.

Sources

  1. Government of Western Australia, Seven golden rules for prospecting (Miner's Right, hand held tools such as a metal detector, samples up to 20 kilograms, fill any holes)
  2. Shire of Cue, Gold Country (each person needs a Miner's Right; metal detecting is prospecting; a Miner's Right is not a mining tenement and does not authorise heavy machinery)
  3. Government of Western Australia, Miner's rights (prospecting includes the use of metal detectors)
  4. Resources Victoria, Golden opportunities to find a nugget this Easter (mechanical equipment or explosives not permitted; a 4.6 kg nugget found with a metal detector in 2023)
  5. Resources Victoria, Fossicking rules and responsibilities (a miner's right is required; electronic detectors are listed among permitted tools; no machinery or explosives; machinery is not defined)
  6. SPH Engineering, Drone based magnetometer system (magnetometers cannot detect gold nuggets)
  7. Coordinates, Maggy, an autonomous magnetometer drone for landmine and unexploded ordnance detection (University of Central Lancashire, April 2025)
  8. Mlambo and others, Inductive metal detectors and the design of prospecting robots: a possibility (EAI, 2018)
  9. Minelab, Gold Monster 2000 launch (a Gold Chance Indicator leans toward probable gold or ferrous targets)
  10. Kitco, live gold price (Nov 16, 2026: $3,861.50 per troy ounce, $124.15 per gram)
  11. Standard Bots, Boston Dynamics Spot robot cost (starts at about $75,000 for the Explorer Kit)
  12. Robotics Center, Spot vs Unitree Go2 (Go2 Enterprise $8,900)
  13. Serious Detecting, Minelab GPZ 7000 ($9,499 sale price)
  14. Earth.com, Man with a cheap metal detector finds a huge piece of gold (most finds are tiny flakes or pea sized pieces)
  15. EarthSky, Opportunity rover (designed for 90 sols, lasted 15 years, traveled more than 45 km)
  16. Machine Design, Martian Opportunity rover RIP (Spirit ended when it got stuck in sand)
  17. Reuters via The Star, Soaring prices spark Australia gold rush (Victoria golden triangle about 9,600 sq km)
  18. Resources Victoria, Golden prospects for Mother's Day (a 119 gram nugget found by two prospectors with a metal detector)
  19. Highbanker, Minelab GPZ 7000, a deep gold getter (nuggets found down to about 1 metre; a retailer blog)
  20. Big Think, Opportunity rover 15th anniversary (45 km and more than 90 million wheel turns)
  21. Government of Western Australia, Prospecting licences (under 200 hectares, must be marked out, up to 500 tonnes of material)
  22. Mining.com, Essential mining laws in Western Australia (an exploration licence can cover up to 21,700 hectares of set blocks)
  23. 43 CFR 3809 text via Idaho Department of Lands (casual use generally includes metal detectors and other battery operated sensing devices)
  24. Bureau of Land Management, Surface management of locatable minerals (casual use is negligible disturbance, with hand tools and no mechanized equipment or explosives; advance contact may be needed where cumulative effects of casual use could exceed negligible disturbance)
  25. Belgian FPS Economy, Deep sea mining, what mineral resources are (nodules grow up to 10 mm per million years)
  26. Discover Wildlife, Clarion Clipperton Zone expedition (4.5 million square kilometres between Hawaii and Mexico; an expedition sailed October 17, 2026)
  27. S&P Global, New Zealand election stirs optimism for seabed mining (December 2023: iron sands off the west coast, vents with gold and other metals, marine consents refused so far)
  28. Te Ara, Marine minerals, nodules, crusts and vents (nodules form in abundance south of New Zealand)
  29. Te Ara, Marine minerals, phosphates and ironsands (Chatham Rise phosphate, about 100 million tonnes, mining risky)
  30. Environment Guide NZ, Minerals (Chatham Rock Phosphate consent application declined)
  31. Dawson City, Klondike Gold Rush (Carmack, Dawson Charlie and Skookum Jim found gold on August 16, 1896)
  32. Government of Yukon, A brief history of placer mining in the Yukon (12 million ounces of placer gold)
  33. Mining.com, Klondike Gold hunts for source of the Yukon Gold Rush (no significant bedrock source found in more than 120 years)
  34. Library of Congress, Siberian mining (Yenisey 1838 to 39, Lena fields, Kolyma under Stalin)
  35. MDPI Remote Sensing, Gold mining land cover change in the Magadan region (Magadan leads Russian alluvial gold, in continuous permafrost)
  36. Australian Antarctic Division, The Madrid Protocol (Article 7 prohibits mineral activity other than scientific research)
  37. World Ocean Review, Mineral resources beneath the Antarctic ice (minor occurrences of gold on the Antarctic Peninsula)
  38. Dark Tourism, Kolmanskop (Lewala, a railway worker who had worked at Kimberley, found a diamond in 1908; about 10 percent of the world market by 1912; peak population about 350)
  39. Travel News Photos Stories, Kolmanskop (hopefuls searched on their bellies on moonlit nights; the Sperrgebiet)
  40. Much Better Adventures, The dark history of Kolmanskop (Lewala not rewarded; forbidden zone declared September 1908)
  41. Allgemeine Zeitung, Tsau //Khaeb National Park (Sperrgebiet declared September 1908; sole mining rights to the Deutsche Diamantengesellschaft)

Inspired by Happy, written by Claude.

Next relatedRadio for Signed Payments: What Ham Rules Allow and What Would Have to Change
Revision history (R3 current)
RevDateWhat changedPage
R0Nov 18, 2026First published. Written by Claude (Sonnet) from Kenny's one line idea and steering in chat, not DeepSeek. Facts were checked against the nine pages listed, and only Western Australia and Victoria were checked. No product that prospects autonomously was found, which is stated as a limit and not as proof none exists. Whether a robot counts as a hand held tool is NOT confirmed by any source. Wikipedia links are from memory. Kenny has not yet said which angle he wants (invention, prediction or warning), so this is the proposal reading.view
R1Nov 18, 2026Added sections on cost, what to expect, the Mars rover idea, shared paths, battery swaps, running at night, monitoring and safety, from Kenny's steering in chat. Eleven sources added. Payback and coverage figures are Claude's assumptions and are labeled as such. Gold, Spot, Go2 and GPZ 7000 prices are mid November 2026 and will change. No armed, fire based, chemical or directed energy defenses are covered, on purpose. Written by Claude (Sonnet).view
R2Nov 18, 2026Added where to search and hold ground (Western Australia tenements, US casual use), depth and the one big nugget, the honest numbers (conservative, rational and aggressive, all my assumptions), other places (Canada, Siberia, Antarctica, Namibia), other sensors and targets, and underwater and power. Twenty one sources added. Gold per hectare is a guess with no data behind it, and the big nugget odds are invented to show the shape. Diamond, lithium, uranium, seawater battery and algae points are general knowledge and flagged. The New Zealand consent facts are from a December 2023 report. Wikipedia links are from memory except Lewala and Stauch, which appeared in search. Written by Claude (Sonnet).view
R3Nov 18, 2026Added Ways to dig, with figures (a person digging, a digging arm on each robot, or one shared purpose built digging robot, all on guessed figures), the Victoria fossicking wording on detectors and machinery, the reading that machinery means digging and ground moving equipment (labeled as a reading, not a ruling), the cumulative disturbance note from the BLM, and the GPS and digging cost caveats. No new sources, two source notes extended. Seven more of Kenny lines are in the arguments box. Written by Claude (Sonnet).view