The caverns hosting the four Large Hadron Collider experiments — ATLAS, ALICE, CMS, and LHCb — are among the harshest environments imaginable for robots. They are confined, cluttered with fragile detector hardware, and entirely denied of satellite navigation. Above all, they are permeated by intense stray magnetic fields spilling out of the powerful magnet systems that surround the interaction points. These fringe fields can disturb onboard electronics, corrupt sensor readings, and destabilise the actuators of any robotic platform sent into the space. Characterising them is a prerequisite for deploying robots in the caverns, but it also matters to detector operation itself: anomalous fringe fields extending beyond the detector volume can interfere with surrounding infrastructure and, if severe, compromise particle tracking and momentum reconstruction.

Conventional approaches to mapping these fields have serious limitations. Purpose-built mechanical systems with Hall-probe arrays work well inside the detector bore but cannot reach the cluttered open cavern. Fixed sensors and manual surveys provide only sparse spatial coverage and require significant access coordination. The robotic airship offers a compelling alternative. Buoyant lift from helium enables extended flight endurance while inherently reducing collision risk, making the blimp well suited for safe indoor navigation near fragile detector components. The platform is built entirely from non-ferromagnetic materials — PLA and DEPRON foam — so the field that it measures cannot attract it toward the massive steel flux-return yoke of the L3 solenoid, the dominant magnetic source in the cavern.

The L3 solenoid is a veteran of particle physics. Originally commissioned in 1988 for the L3 experiment at the Large Electron–Positron collider and later adapted for ALICE, it generates a nominal axial field of 0.5 tesla over a volume roughly 12 metres long and 11.8 metres in diameter, driven by 30,000 amperes through 168 turns of an octagonal aluminium coil and storing 150 megajoules of energy. A warm forward dipole magnet, providing 0.67 tesla for the muon spectrometer, sits nearby. The fringe field extending beyond the steel yoke into the cavern — typically in the millitesla range at the measurement locations — was the primary target of the blimp’s campaigns.

The measurement payload is elegantly simple. A TMAG5273 linear three-axis Hall-effect sensor from Texas Instruments, operated in the millitesla range and sampled at 10 hertz, provides the magnetic readings. It is read by an Arduino Portenta H7, which also acquires an inertial measurement unit, environmental sensors, and a time-of-flight proximity sensor. Because every field measurement must be tied to a spatial position, a Raspberry Pi Camera 3 Wide, operated by a dedicated Raspberry Pi Zero 2W, streams images used for localisation. The two data streams are timestamped independently and synchronised via Network Time Protocol, limiting temporal misalignment between field readings and pose estimates during post-processing.

Localisation is the crux of the system, since the cavern offers no satellite navigation. The team deployed 82 ArUco fiducial markers — square binary patterns designed for robust visual detection — printed on continuous paper rolls and arranged in strips along the joint lines between the concrete shielding blocks in front of the L3 solenoid. A JSON vocabulary file associates each marker with its three-dimensional coordinates in the ALICE reference frame, derived from the official detector CAD model. The onboard camera detects visible markers, solves the Perspective-n-Point problem to estimate its pose, and chains transformations through the camera-to-sensor offset to express the Hall sensor’s position and orientation in the ALICE frame. The measured field vector is then rotated into the same frame for direct comparison with the simulation.

Two complementary campaigns were conducted with the magnets at nominal current. The first consisted of static measurements: the Hall sensor was mounted on a 3D-printed platform with the same footprint as an ArUco marker, placed at twenty known positions in two parallel strips, and a 30-second acquisition was triggered at each point. By eliminating localisation uncertainty, this campaign isolated the Hall sensor’s accuracy from the pose-estimation error. The second campaign consisted of four manually piloted flights on two separate days in 2026, during which the blimp acquired measurements continuously while navigating over the same instrumented region, exploring a volume of several cubic metres in front of the A-side shielding blocks.

To compare measurements with the electromagnetic model, the team built a high-fidelity simulation of the ALICE magnet system in CST Studio Suite, including the L3 return yoke and forward dipole yoke with the nonlinear B–H curve of AISI 1050 annealed carbon steel, and generated a field map sampled on a uniform 200-millimetre grid. A dedicated uncertainty-propagation framework then combined the Hall-sensor error budget — sensitivity error, zero-field offset, and intrinsic RMS noise — with the ArUco pose uncertainty, which was characterised in a controlled indoor environment against a Qualisys motion-capture ground truth. The validation campaign yielded a three-dimensional position RMSE of 5.38 centimetres and attitude errors within about 2.6 degrees, with the Hall-sensor contribution dominating the total uncertainty in every flight.

The results are strikingly consistent across both campaigns. The simulation reproduces the transverse field components well: statically, the x-component achieved a coefficient of determination of 0.98 with 100 percent of simulated values falling within the sensor uncertainty bounds, and the y-component reached 0.94 with 90 percent compliance. Dynamically, pooled over 858 valid points across the four flights, the transverse components showed coefficients of determination of 0.86 and 0.93 with compliance of 89.4 and 79.3 percent. The vertical component, however, stood out as the clear outlier in both campaigns, with a lower coefficient of determination of 0.75 to 0.79 and a systematic positive bias of 0.65 to 1.12 millitesla — the measured field consistently higher than the simulation predicted.

Critically, because this vertical bias appears already in the static campaign, where the ArUco localisation contribution is entirely absent, the authors conclude that the discrepancy originates primarily in the electromagnetic model rather than in the measurement pipeline. Two modelling limitations are the most plausible culprits: the actual yoke materials are not documented, introducing systematic uncertainty in the field components most sensitive to the permeability value, and the CST geometry does not include the ferromagnetic concrete shielding blocks adjacent to the solenoid access doors, which may locally perturb the field. A parametric sensitivity study varying the yoke permeability and shielding geometry is planned as future work. The team also notes that the current system depends on marker visibility for absolute localisation, and that the four manually piloted flights demonstrate the methodology within an instrumented test region rather than a complete cavern survey.

The significance of the work extends well beyond ALICE. The High-Luminosity LHC will employ stronger focusing magnets and further restrict human access, and the proposed Future Circular Collider foresees no iron return yoke for the solenoid flux at all — under that assumption, the stray field at the cavern wall is expected to reach approximately 100 millitesla, falling below 5 millitesla only beyond 55 metres from the detector. Such conditions, orders of magnitude more intense than in current LHC caverns, make magnetic compatibility a critical design driver for any future autonomous inspection platform. The next step for the blimp is autonomy: integrating the ArUco localisation pipeline with a closed-loop position controller and a field-aware path planner would allow it to execute pre-programmed survey patterns, repeat missions consistently, and function as a continuous monitoring system capable of detecting field anomalies during beam operations without requiring human presence in the cavern. The measurement methodology, localisation pipeline, and uncertainty framework developed in this work constitute the foundation for such a fully autonomous airship-based survey system in underground physics environments.

Subject of Research: Robotic airship for measurement and validation of a CERN detector stray magnetic field

Article Title: Robotic airship for measurement and validation of a CERN detector stray magnetic field

Article References: Mazzei, F., Scaramuzzino, P. F., Bernardi, L., Curti, F., & Gargiulo, C. (2026). Robotic airship for measurement and validation of a CERN detector stray magnetic field. Frontiers in Robotics and AI, 13, Article 1926428. https://doi.org/10.3389/frobt.2026.1926428

Image Credits: AI Generated

DOI: 10.3389/frobt.2026.1926428

Keywords: Robotic, airship, measurement, validation, CERN, detector, stray, magnetic, field, scientific research