Confined space inspection robots in the UK: which robot suits which space, and what the law asks first

For inspection, often. For the work itself, rarely. Regulation 4(1) of the Confined Spaces Regulations 1997 says that no person at work shall enter a confined space to carry out work for any purpose "unless it is not reasonably practicable to achieve that purpose without such entry". The Health and Safety Executive puts it more plainly: wherever possible, avoid carrying out tasks in confined spaces. Inspection (looking, measuring, recording) is the task most readily done without entry. Cleaning, repair and clearing blockages usually still need people or specialist tooling.

The Regulations define the risks that make a space "confined": serious injury from fire or explosion; loss of consciousness from raised body temperature; loss of consciousness or asphyxiation from gas, fume, vapour or lack of oxygen; drowning from a rising liquid; and asphyxiation or entrapment by a free-flowing solid such as grain or sand. None of those risks applies to a machine in the same way, which is why a robot changes the order of work. A person goes in when the robot's evidence says there is a job to do, not to find out whether there is one.

UK operators have published early results. UK Power Networks, working with Arup under Ofgem's Network Innovation Allowance, trialled a four-legged robot in cable tunnels and shafts and estimated the approach could cut staff time in high-risk locations by up to 50%. Network Rail used a small legged robot to inspect a culvert under the railway in Essex in February 2025 without staff entering the structure. Those are the operators' own statements and estimates, not Forge results.

There is no single confined space robot. The table describes robot forms, not named products; the pipe crawler column covers what suppliers usually list as pipe inspection robots. Whether a given platform suits a given space depends on its published dimensions, ingress rating, certification and payload, checked against a survey of the space itself.

The Confined Spaces Regulations 1997

Three duties matter. Avoid entry where reasonably practicable (regulation 4(1)). Where entry cannot be avoided, work to a safe system (regulation 4(2)). Put emergency arrangements in place before anyone enters (regulation 5). HSE's Approved Code of Practice, "Safe work in confined spaces" (L101), explains how those duties are met. Using a robot does not stop a space being confined. If the robot fails inside and someone has to fetch it, that recovery is confined space work with its own permit, atmosphere testing and rescue plan. Planning how a stuck robot will be recovered, by tether, recovery line or a second robot, belongs in the method statement before the first deployment.

Gas, dust and explosive atmospheres

Where a flammable gas, vapour, mist or combustible dust atmosphere may occur, the Dangerous Substances and Explosive Atmospheres Regulations 2002 (DSEAR) require the area to be classified into zones, and equipment in each zone must be of the matching category: category 1 in Zone 0 or 20, category 1 or 2 in Zone 1 or 21, and category 1, 2 or 3 in Zone 2 or 22. A robot is equipment. Most inspection robots on the market are not certified for any zone and must stay out of zoned spaces. Our guide to ATEX and hazardous-area inspection robots in the UK covers which forms are certified and what the marking means.

Sector rules that sit alongside

A quadruped earns its place in confined spaces that were built for people to walk through: cable and service tunnels, large culverts, plant rooms, basements, undercrofts and pump houses. It can run the same route with the same viewpoints on each visit, carry thermal, zoom, acoustic and gas sensors, and walk steps and rubble that a wheeled robot cannot. The limits are physical and worth measuring before any platform is chosen.

The robot is the carrier; the sensors do the inspection. For atmosphere, a gas payload matched to the space, with oxygen, flammable gas (lower explosive limit), hydrogen sulphide and carbon monoxide as the usual minimum, fitted, rated and calibrated. For condition, a zoom camera with its own lighting, because confined spaces are dark. A thermal camera for cables, joints, bearings and hot spots. LiDAR where the question is deformation, silting or change over time. Ultrasonic thickness where the question is metal loss, which is crawler and contact-drone work. Sonar where the space is flooded. Each payload adds weight against the robot's rating and must itself be rated for the environment.

A robot's gas readings are useful evidence of the conditions it saw. They do not replace the atmosphere test that the permit requires for any person who then enters, at the time they enter.

A confined space inspection sits in a chain of records: the permit, the inspection record, the defect, the work order and the asset register. Most robots report into their maker's own app, so someone has to notice a finding there and type it into the system that owns the work. That is where deployments stall.

Forge Anvil is the integration software we are developing for this job: getting what a robot finds into the systems your team already works from. For confined space inspection, the pattern we would scope looks like this:

Which systems Forge Anvil connects to is scoped site by site in the integration review. How the four integration patterns compare, and what to demand from a manufacturer, is in our guide to integrating a robot with the software you already run.

Start with one type of space and one inspection. Survey the access first: the narrowest point, the headroom, the steps, the water and the signal. Choose the robot form from the table above, then the platform from its published ratings. Involve the people who sign the permit and hold the rescue plan from the first meeting, not after the robot arrives.

Agree the measures before anything goes in:

Set exit criteria too. If the engineer will not accept the evidence, or a recovery needed entry, the pilot stops and says so. That is how we structure a pilot programme, on a manufacturer-neutral basis, and a robot for the pilot can come through a demonstration, a hire or loan, Robotics-as-a-Service where a maker offers it, or supply through Forge where that fits. The trial routes are compared in how to trial a quadruped robot in the UK without buying one.

Grant or innovation funding can sometimes contribute towards a suitable pilot. Eligibility depends on location, sector, applicant type, project stage and the schemes open at the time, approval is at the funder's discretion, and funding is never guaranteed, so it is assessed before a pilot is scoped rather than assumed into the budget.