Telerobotics · Sensing

A rover for georeferenced radiation mapping

Commercial radiation telerobots are priced for national laboratories. This one is a Raspberry Pi, a Geiger tube and a 3D-printed chassis. Surveying around Chapelcross, it found a band of elevated count rate that matches a buried effluent pipeline the 2015 site survey doesn't record.

Context: MEng Individual Project, Southampton
Role: Sole author, whole system
Period: 2020 to 2022
Field sites: Chapelcross, Southampton
Report: Full report (PDF)

Plan view of the rover powered up: four large green 3D-printed tyres on black printed suspension arms with blue anodised dampers, a lithium-polymer battery strapped over the chassis, a brushless motor and speed controller with cooling fan, and an inline current meter reading 0.07 amps. — The rover under power. Largely 3D printed, built around the open-source Tarmo radio-control chassis design, carrying the detachable sensor unit, a Raspberry Pi, GPS, camera and battery.

35–52%elevation in count rate along the buried pipeline

28%of the cost of the commercial equivalent

ρ = 0.76correlation with that commercial unit over 794 readings

The problem

Radiation is difficult to survey. There is nothing to see or smell, and the areas most worth measuring are the ones where people should spend the least time, so the work suits a telerobot. The equipment that does it properly is beyond the budget of a local authority, a university group or anyone wanting to check a site independently.

The aim was a radiation monitoring telerobot a person could actually build: a mobile platform under £150 of parts, a sensor unit under £200, and an output a non-specialist could read. The whole thing followed an open-source methodology so somebody else could rebuild it or point it at a different measurement entirely.

The rover chassis on a cutting mat during assembly, with a Raspberry Pi GPIO pinout reference card beside it and servo wiring routed through the frame. — Mid-build. The steering servo and the sensor bay were the two constraints that drove the chassis layout.

The sensor unit, a Raspberry Pi and Geiger board in a printed vented case with ribbon and jumper wiring, photographed next to a domestic smoke detector with a five centimetre scale bar. — The sensor unit beside a domestic smoke detector, which is also the americium-241 source used to confirm the detector responds to anything at all.

The rover and the logger

The design settled into a rover and a detachable georeferenced radiation logger that rides on it. Separating them mattered for two reasons. The logger is the part that has to be trustworthy, and it can be characterised on its own. Because it detaches, it can also be carried in a backpack or fixed to a car when a survey covers open ground.

Count rate and dose-rate estimate logged continuously, every reading tagged to a GPS fix.
Teleoperation with a Bluetooth games controller, relayed over 4G through a Raspberry Pi Zero, so range isn't bounded by radio line of sight.
An operator GUI written in tkinter, carrying a live first-person video stream alongside current count statistics, which makes driving at distance workable.
Interpolated heat maps as the output, plus interactive maps for inspecting the underlying readings.

Latency and standoff

Driving a vehicle you can't see, through a video feed, falls apart quickly if the delay between input and picture is long enough to notice. The first working version was well over that threshold, and most of the engineering effort went into the optimisation that followed, which brought latency down to 0.11 s. Past that point the operator no longer has to compensate for it, and the rover was driven at distances exceeding 300 metres.

Driving from the interface rather than by eye. Live video on one side, count rate on the other, both over a mobile network.

Checking the detector responds

Before any mapping is worth doing, the detector has to be shown to react to a source. A domestic smoke detector contains a small sealed americium-241 element, which makes it a convenient known quantity to move towards and away from.

Count rate against time over 380 seconds. The rate sits around 18 counts per minute background, rises steadily from about 90 seconds to a peak of 43 counts per minute at 139 seconds, then decays back towards background. — Count rate through an approach and withdrawal. Background sits around 18 CPM and the peak reaches 43. The raw trace is noisy enough that the rolling mean matters, which is the central constraint of a cheap tube: count statistics need dwell time, so survey speed trades directly against confidence.

Into the field

Chapelcross was the obvious site. One of the earliest British nuclear power stations, running Magnox reactors from 1959 to produce both electricity and weapons-grade plutonium-239, in decommissioning since 2004 with work expected to continue past 2095, and now home to the last radioactive waste pond in Scotland.

The platform had to be robust enough to take repeatable readings while crossing rough ground, which is what most of the site consists of.

Driving over woodland leaf litter and roots.

What the operator sees. The forward camera at ground height.

Logged readings interpolated into a readable surface.

I surveyed roughly three square kilometres around the perimeter, and separately around Southampton for a baseline of ordinary background.

Map of the area around the former Chapelcross nuclear power station with the survey route outlined in black and an interpolated count-rate field overlaid in red and blue. — The wide survey. The outlined corridor is the route actually driven and the shading is the interpolated count rate along it. The fill has to be read with the route in mind, because it interpolates across ground the rover never covered.

The pipeline

One area came back consistently warmer than its surroundings, so I went back and mapped it properly: 3,600 square metres, a reading every four seconds, walking adjacent lines five metres apart, using the interface to hold a constant line of latitude so the passes stayed straight.

The result was a band rather than a patch, running diagonally across the survey area at count rates 35 to 52 per cent above the surroundings. Contamination doesn't usually deposit in a straight line, so the shape pointed at something buried underneath it.

A close-range interpolated count-rate map of a small area, with latitude and longitude axes. A distinct diagonal band of elevated count rate runs from the lower left to the upper right against a lighter background. — The detailed survey. The diagonal band is what prompted the search for something buried underneath it.

The 2015 survey of the site doesn't document anything on that line. A 2002 Radiological Habits Survey does, in passing: an underground effluent pipeline whose route coincides with the band.

The corroboration is imperfect. The 2002 survey used a gamma-only counter, so its readings are not directly comparable with the sensor unit's. What it does provide is a trend, a 60 per cent reduction in dose rate moving south towards the point where the pipeline crosses the A75, and that trend appears in my data too.

Validation

Validating a cheap instrument needs a better one to compare against, and without a dosimeter the absolute accuracy of the Geiger counter could not be established directly. What could be established was agreement and repeatability, which is most of what a survey map depends on.

Against a comparable commercial unit, across 794 collected data points, the readings correlated at a Spearman's rank coefficient of 0.76. Precision came out at ±7.3 per cent of recorded dose rate, and the sensor unit placed a reading within 8 metres of its true position 95 per cent of the time. The combined build cost across both devices came in inside the £350 target, at around 28 per cent of what the commercial offering costs.

It isn't accurate enough to certify a site as safe. It can show where a proper survey should look.

← All projects