Recently, a team from the National Polytechnic Institute has presented a Exploratory robot capable of mapping mines and identifying risks, combining neural networks and computer vision to improve safety in underground environments.
The development is signed by the students Carolina Abigail Gallo Meneses, Yesenia Cruz Dominguez, and Lesly Veronica Salazar Jimenez, from UPIITA (Telematics Engineering), with academic advice, and is oriented to Prevent accidents through early detection of landslides, fractures, and floods, in addition to toxic gases.
Robot technology and design
The prototype is based on a commercial exploration vehicle to which a High-performance Raspberry Pi 5, a lamp for local lighting, a depth camera and two specific sensors for carbon monoxide (CO) and nitrogen dioxide (NO2).
Thanks to the depth camera, the rover captures RGB images even in total darkness and feeds a simultaneous mapping and localization system (Visual SLAM, V-SLAM). This combination enables navigation in tunnels without signal and accurately recreates the surroundings.
During exploration, the robot generates a preliminary map with points and boxes as spatial references. Subsequently, and already outside the mine, processing at the base station transforms this data cloud into detailed three-dimensional models of the work face, where fractures, accumulations of stones, areas with landslides and flooded areas can be distinguished.
3D mapping and offline processing
The flow is designed to operate without a network: V‑SLAM works locally during the journey, and once on the surface, heavy post-processing is performed. This architecture reduces operational risks and avoids reliance on connectivity or GPS, which are nonexistent. in deep sections of the galleries.
The team reports that, dozens of kilometers from the entrance - sometimes it is said about 30 kilometers inland—, satellite positioning is completely lost. For this reason, the platform was designed to be fully autonomous and functional offline.
AI training and risk detection
To fine-tune the neural network, we started from some 5.500 initial images, which after augmentation techniques (rotations and position variations) climbed to approximately 13.000 samplesThe set comes from both a mine in Durango and a test model built for the project.
With that training, the system learns to recognize geological hazard patterns and correlate environmental conditions with gas readings, which helps prioritize areas of attention and implement more effective security protocols.
Field tests and mine operations
The creators made a technical visit to a mine in the state of Durango to observe working conditions in situ and validate the rover's performance. Comparison with foreign solutions highlighted a key difference: many available technologies are Fixed systems that require manual movement of modules, while this proposal is mobile and autonomous.
This mobility, combined with 3D mapping and CO and NO2 measurement, makes it easier quick inspections without exposing personnel, especially in galleries with reduced visibility, high humidity and potential presence of dangerous gases.
Web platform and visualization for decisions
In addition to the hardware, the team developed a web system for storing and displaying information collection. Includes navigable three-dimensional maps, geospatial location associated with the route, the time of exploration and graphs with the gas measurements.
This interface allows security officers and technical teams interpret data at a glance and support operational decisions, from the temporary closure of a sector to the planning of structural reinforcements or ventilation.
Institutional momentum and next steps
The project is aligned with public policies of technology transfer: it is part of the commitment 33 of the 100 presented by the Presidency and promoted by the SEP, so that student developments can make the leap from the classroom to the industry.
He had the advice of Dr. Rodolfo Vera Amaro (UPIITA) and the Dr. Lucero Verónica Lozano Vázquez (ESIME, Azcapotzalco Unit). The team does not rule out manage a patent and continue to refine it for large-scale industrial uses.
It is a solution that integrates V‑SLAM, machine vision and gas sensors in an autonomous vehicle, with post-processed 3D models and a web analysis layer, aimed at raising the mining safety with objective evidence and without depending on field connectivity.
