Researchers from Carleton University and MDA Space Missions at the CANMET Experimental Mine in Val-d’Or, Quebec, where they recently carried out field testing of some concept robotic mapping and underground positioning technologies (left to right: Jamie Lavigne, Unal Artan, author Joshua Marshall and Stefan Radacina-Rusu)
In surface mining, the introduction of the satellite-based global positioning system (GPS) has had a remarkable impact, possibly representing one of the most sweeping technological revolutions in the past 30 years. Together with advances in wireless data communications, GPS has made material tracking and fleet management systems nearly commonplace in surface operations.
Currently, however, no directly comparable technology exists for accurate and real-time positioning of mobile equipment underground. Yes, there are partial solutions, but a breakthrough product available to mining companies that operate underground has not arrived.
Because satellite GPS signals cannot usefully penetrate obstacles like rock, one practice is to place many uniquely identifiable radio frequency devices (e.g., RFID tags/receivers or Wi-Fi access points) throughout the mine. This method requires that the wireless devices be placed at known locations so that equipment can be roughly monitored by registering sensed devices with a central database of device locations by way of an underground communications network.
Although this method may be sufficient for some purposes (such as entry-point tracking), it lacks many of the advantages that the satellite GPS provides on surface. Underground, only occasional positioning information is provided — whenever a vehicle happens to be near a device. Installing large numbers of tags for more precise positioning is not an elegant solution.
Positioning truth from fiction
Unfortunately, common misconceptions exist about the state-of-the-art in underground mapping and vehicle positioning. The first is that radio signals (e.g., Wi-Fi signals) can directly provide GPS-like positioning in underground mines. There have been some promising advances in this area, but even the most recent research suggests that in non-trivial environments — where there are obstacles that prevent direct line of sight to all devices — the problem has yet to be fully solved. Three or more overlapping radio signals are nominally required, which is not really practical for underground mines, and problems such as severe multipath, shadowing and interference have plagued developers. Nevertheless, this approach does show promise.
A second misconception is that inertial measurement units (IMUs), which consist of accelerometers and gyroscopes, can alone be used for global positioning and mapping. This method is commonly known as “dead reckoning” and works by integrating motion measurements over time to estimate position and orientation. However, measurement errors are also integrated, resulting in unbounded estimation uncertainty over time. Early naval vessel navigators knew this very well. When used for mapping, dead reckoning makes it impossible to determine with any accuracy when a vehicle has returned to a previously traversed area.
Efforts to improve IMU measurements with visual data (e.g., from lasers or cameras) have a long history in the robotics community, such as in planetary rover navigation and mapping, yet the error accumulation problem often remains. There are some very accurate IMUs on the market, but their high cost makes them impractical. And, ultimately, estimation uncertainty still grows, just more slowly.
Another misunderstanding is that the technology exists for accurately triangulating underground position by using rock-penetrating very low frequency (VLF) signals. VLF signals have a wavelength of between 10 and 100 kilometres. To the best of our knowledge this has yet to be demonstrated and thus, at present, we rate this possibility as inconclusive.
Finally, some believe that existing autonomous mining vehicle products already solve the robotic mapping and positioning problems. Yes, the industry has seen great success in the development of autonomous tramming technologies (although, as a designer, the author may be slightly biased in this regard); however, these systems have not explicitly required a solution to the positioning problem with dead reckoning. This is because they are in fact trained to follow a particular route irrespective of whether the route crosses itself or not. In other words, the nature of the autonomous tramming task requires that the vehicle perform only local positioning, not fully global positioning.
While advances in these areas may be realized in the future, technology consumers should be alert to the difference between credible advances and what may simply turn out to be wishful thinking.
Map-based global positioning
Suppose that we could create an accurate “map” of an underground drift network, one detailed enough to include even bumps in the walls and other pertinent information. Could we devise a technique for estimating a vehicle’s position and orientation within that map? Would this be good enough? Recent developments in robotic mapping and map-based underground positioning by researchers, including the author, from MDA Space Missions in Brampton, Carleton University, and now Queen’s University, funded in part by the Ontario Centres of Excellence and the Natural Sciences and Engineering Research Council of Canada, has focused on this question.
Such a system would hinge on the ability to create an accurate and globally consistent map. As described earlier, if only dead reckoning were used to build the map, then inconsistencies in the map would inevitably occur. This would be especially apparent in a large environment, such as an underground mine level. The resulting map would be “inconsistent” because it would contain multiple points on the map that represent the same point in the actual environment. This would occur when the sensors used to acquire mapping data return to an already
In this work, we have concentrated on the problem of creating a planar grid map — also called an “occupancy grid” — of a large-scale underground mine by using dead reckoning combined with a scanning laser rangefinder. First, information about the environment is collected, which is then used in an offline map generation process. We have been successful at solving the map consistency and positioning problems with an optimization approach that relies on matching laser scans.
Preliminary field testing has taken place in various environments during the past year as we now approach the objective of delivering a GPS-like underground navigation product for underground mining. There is growing confidence that our system will have the potential to improve safety, contribute to the advancement of remote and automated vehicle systems, and enable the design and deployment of unique and truly useful asset management systems for underground operations.
Joshua Marshall, P.Eng., is an assistant professor in The Robert M. Buchan Department of Mining at Queen’s University, where he studies advanced mining systems and robotics. He is also an adjunct research professor in the Department of Mechanical and Aerospace Engineering at Carleton University. He holds two patents related to mining machine automation and has published on the subjects of mobile robotics and equipment automation.