Magnetic mapping and navigation
You don’t need a map to walk from your living room to your bedroom — or is that really true? If you find yourself in a new house, you must first explore the space to determine the layout of the rooms using various cues in the environment such as walls, furniture, lights, etc. Once this map is “created,” you will be able to plan your movements without any second thought. As a general rule, the more senses are available to build this map, the more precisely and reliably will you be able to navigate — but you do not need all your senses to do so. You can likely close your eyes and walk through your house without getting lost — perhaps with the additional help of your arms to reach for walls!
Mimicking such intuitive human behavior is exactly what one would aim to achieve with autonomous robots. To give a concrete example, GPS can be used to position a robot on the Earth in the open outdoors as it is now cheap and available almost everywhere. That said, this “sense” breaks down the instant the sky is obstructed — which happens inside buildings, in underground tunnels or even in forests with dense tree canopies. The internal navigation system of the robot (typically composed of wheel odometry sensors and an Internal Motion Unit sensor) can take over but only for brief periods of time as this solution is relative. Indeed, such sensors measure the internal state of the robot without any external reference, making errors difficult to detect and correct — which keep on adding up over time.
This is where an alternative “external” sense is required to fill-in the role of GPS for absolute positioning — common optical solutions like LIDAR or novel applications of magnetism for instance. Akin to navigating around the Earth with a compass, the local patterns in the magnetic field of the Earth change for different latitudes and longitudes so they can act as markers for positioning and navigation [1]. The main advantage here is that the magnetic field can be measured everywhere — including underground, underwater, in the air and even in space.
A further advantage is that the Earth’s (mostly) dipolar magnetic field is not alone in shaping the magnetic map. For example, buildings typically consist of a complex arrangement of steel rebar, studs and other magnetically active objects that all contribute to the local magnetic field. Such features can act like “beacons” or “lighthouses” that can be used to navigate nearby or unambiguously identify vicinities; for example, sensing the magnetic signature of a fridge or hearing its familiar humming are both tell-tale signs that one is in or close to the kitchen! Another advantage here is that such signatures are complex and cannot be “beamed” towards a receiver in the manner that GPS signals can be spoofed [2]. Such attempts would virtually be impossible except with extremely precise knowledge and control of the environment close to the sensors.
How does this work in practice? Suppose a given position has known magnetic information; recording that magnetic information is a pretty good clue that you might be located exactly at that given position. This is a guess of course as you might imagine: what if two positions share the same magnetic data? In such cases, one relies on the path traced so far to establish which of the choices is more likely. Such details are handled by sophisticated mathematical algorithms using statistical principles, but the fundamental idea is that of a (reverse) lookup table.
The notion of magnetic information is subtle yet extremely important — but what is clear is that the more information is available, the more detailed and reliable the map can be. In our previous blog post, we discussed SBQuantum’s involvement in measuring not only the magnetic field vector, but also the magnetic gradient — combined, these provide eight maps that can be used as sources of information.
Now the management of this magnetic dataset as well as the input from other sensors on an autonomous robot can be integrated in a mathematical framework called Simultaneous-Navigation-and-Mapping (SLAM) [3]. As its name implies, such algorithms attempt to condense all the sensor data to produce estimates for both the map itself as well as to localize the robot in it at the same time. This may seem quite paradoxical as one would expect the map to be needed first before navigating, but the two problems can in fact be treated as a single optimization problem with some clever engineering. Once solved, the result is a map and a trajectory that best explain the data registered by the robot over the last time interval.
SBQuantum has partnered with Professor James Richard Forbes and M.Sc. student Natalia Pavlasek from the DECAR lab at McGill University to build towards magnetic SLAM algorithms [4]. Such algorithms combine SBQuantum’s expertise in handling magnetic gradiometry information and the mathematical and robotics expertise of the DECAR lab. This solution improves on previous attempts to adapt the SLAM framework to magnetism by including what are known as “loop closure” constraints via magnetic invariants, the measurements of which are less affected by attitude errors.
Autonomous navigation is a very exciting field of robotics that has come leaps and bounds in the last few decades and must now be made precise and robust in ever increasing complicated scenarios. Of particular concern is overreliance on GPS, especially in defense scenarios. Moreover, current autonomous vehicle solutions work well in the simple environments of southern US states, but perform poorly in Canadian winter conditions or the traffic chaos of many Southeast Asian cities. While still in early stages of development, magnetism is another sense that robots can and should leverage to fill-in the critical “blind spots” of other technologies. At SBQuantum we strive to produce a magnetic intelligence solution that makes it viable not just for primitive compassing, but also as a full-fledged autonomous navigation solution.
[1] A. Canciani and J. Raquet, “Absolute Positioning Using the Earth’s Magnetic Anomaly Field,” in J Inst Navig, vol. 63, pp. 111–126, 2016, doi: 10.1002/navi.138.
[3] H. Durrant-Whyte and T. Bailey, “Simultaneous localization and mapping: part I,” in IEEE Robotics & Automation Magazine, vol. 13, no. 2, pp. 99–110, 2006, doi: 10.1109/MRA.2006.1638022.
[4] N. Pavlasek, G. Martine-La Boissoniere, Z. Flansberry, Thomas Hitchcox and J. R. Forbes, “Simultaneous Localization and Mapping using Attitude-Invariant Magnetic Field Information for Loop Closure.” Submitted.
