There’s no denying that society is trending towards an autonomous future. From delivery robots to drones to self-driving cars and everything in between, the tech industry is hyper-focused on developing solutions that make our lives easier and more convenient. Navigation plays a huge role in this development, as most autonomous products involve getting a person or object from one place to another without any human effort.
Despite how integral navigation is to the development of autonomous solutions, it’s actually a pretty divisive topic in the tech community. No one doubts the importance of powering robots, drones, or AVs with navigation capabilities, but two different schools of thought exist on how exactly to do so. Generally speaking, today’s autonomous tech developers tend to favor either absolute or relative positioning – and have VERY strong feelings about it.
Wait – what’s the difference again?
Absolute navigation is the method of determining where you are and where you are going using precise geospatial coordinates. Essentially, every object has a geographic location on Earth’s surface, and those location coordinates are used to navigate between different places and objects. Consumer mapping applications are an example of absolute navigation; if you type in two addresses, a specific route will be calculated based on the geographic coordinates of those places and the locations of road networks connecting them.
Absolute navigation relies on precise location coordinates to determine the best route.
Relative navigation involves using perception of immediate surroundings to get from one place to another, and is typically best for obstacle avoidance. This is how humans navigate in the world; for example, if you are walking around in your home in the dark, you will rely on your own knowledge of where furniture is located, as well as anything you can see or feel around you.
Relative navigation relies on sensing nearby objects to determine the best route.
There are benefits and limitations to each of these navigation methods, making them more or less ideal for some applications than others. However, when it comes to developing autonomous navigation systems, we strongly believe that the only way to minimize risk and maximize precision is to combine the best of both absolute and relative capabilities.
If you’re familiar with this space, you probably already have an opinion on the relative vs absolute navigation debate. To show you why just one method is not enough, we’ll break down what an autonomous solution could look like in each of these three scenarios: powered by absolute location, powered by relative position, and powered by just the right combination of the two.
The case for absolute navigation
Let’s start with absolute navigation. In theory, this is the perfect solution. Computers have perfect recall, meaning they can store and retrieve the geographic coordinates of any person, place, or object. These coordinates can hypothetically be recorded for everything in the world, generating an absolute frame of reference for all possible destinations and obstacles. Humans could never remember the exact coordinates of everything in the world, let alone calculate the precise degrees and directions of movement required to navigate amongst them. Supporters of absolute navigation say we should take advantage of this dichotomy and allow computers to do what humans cannot.
But is anything ever really perfect? No. Geospatial data is incredibly difficult to keep up-to-date with our dynamically changing world, so no true absolute frame of reference exists at any given moment. Sure, there are datasets that come close using the latest in GNSS positioning tech, but they are not equally distributed throughout the world. The truth is navigation still requires some level of nuance.
Take for instance an AV driving down a road. Even though it’s equipped with location data representing the lines on the road, the bike lane next to it, and the trees just beyond the curb, it does not have the coordinates of the vehicle that just abruptly stopped five yards ahead. In this instance, a human or system perceiving the stopped car would be better able to avoid a collision.
The case for relative navigation
So if perception and reasoning is superior to absolute navigation, then shouldn’t we only develop autonomous solutions using relative location? Not necessarily. While computers have perfect recall, they still do not have the level of fidelity in interpreting the environment that humans do. Computers may surpass humans in many things, but contextual interpretation is not one of them (yet).
There are a few reasons why computers cannot decode the real physical world with the same reliability as humans. Our world is not only changing incredibly fast, but human perception is also highly localized based on the varying geography of different places. Weather is another great example of dynamic conditions that require perception-based judgment. There’s no way for a computer to anticipate and plan for every possible real-world scenario, so precise location alone cannot provide all of the answers. That being said, relative navigation does have limitations, mainly when it comes to encountering situations with no known reference points to contextualize.
Let’s imagine the AV again. The vehicle’s relative navigation system is using the double yellow line to determine how to stay in its lane, but then an intense fog rolls in that completely obscures any road markings. Without precise geographic coordinates representing the double yellow line, the vehicle does not have a way to perceive its own location or stay in the appropriate lane.
The reality: we need both absolute & relative navigation for autonomous solutions
All of this means that in the great debate over relative vs absolute positioning solutions, no one is really 100% in the right – or the wrong. There are definite benefits and limitations to each method, and one alone may be sufficient depending on your use case. However, the high-stakes nature of autonomous navigation means we cannot simply accept the drawbacks of one solution without trying to mitigate them with the other. That’s why we argue that the future of autonomous navigation lies in a unique combination of both absolute and relative techniques.
The reality is autonomy is not binary, and safety is a gradient. Everything in autonomous tech development should be thought about in terms of allowable risk. Redundant navigation systems allow us to reduce risk in scenarios that would be too dangerous to proceed in using just one system alone. The trick is optimizing both absolute and relative navigation to produce the most safe and most autonomous system possible.
We need to introduce human- or computer-based countermeasures for situations when one navigation system fails. The probability conditions of absolute and relative navigation help each system make critical decisions, but their combined joint probability is always going to be greater than that of either method on their own. In other words, AVs, robots, drones, and other tech leveraging autonomous navigation should utilize the best of each method, optimizing for safety, accuracy, and the highest degree of autonomy possible.
The safest and most accurate way to navigate around the Earth is to combine relative and absolute positioning.
Combining absolute & relative navigation
At this point, you’re probably thinking that this all sounds great in theory, but is next to impossible in practice. Configuring and maintaining just one navigation system is resource-intensive, let alone two. But new innovations in GNSS technology are changing the game for autonomous navigation developers, making it not only possible to leverage the strengths of both absolute and relative navigation, but preferable.
AVs and other autonomous tech are almost always already equipped with the basic hardware needed to enable redundant navigation systems. GNSS sensors are ideal for powering absolute navigation, calculating precise positioning based on its existing reference frame of location coordinates; cameras serve as relative navigation sensors, using a visual odometry pipeline to perceive its surroundings and influence movement. When these systems are run simultaneously, their signals can be analyzed for any discrepancies, and the autonomously navigating object can use logic to determine which to rely on in a particular scenario. The data collected through relative positioning can even then be used to inform the absolute reference frame, continuously improving the overall navigation system and keeping it up-to-date with real-world change.
Using our previous example of an AV encountering an abruptly stopping car and foggy conditions, let’s see how redundant absolute and relative systems can work together for optimal autonomous navigation. When the vehicle five yards ahead stops unexpectedly, the relative sensors can overpower the absolute navigation system to inform the AV to also stop. When the AV encounters a foggy area that obstructs the relative navigation system, the reference frame of the absolute system can take over to ensure it stays within the known lanes.
You may be thinking of all the ways these systems can contradict each other. What if the car abruptly stops in the fog, obstructing the relative navigation sensors, but is still not represented in the absolute reference frame? That’s a valid point – like we said earlier, nothing is perfect. But the combined probability of absolute and relative navigation systems is always going to be more resilient than either system working perfectly on its own.
So why isn’t this the status quo in autonomous navigation? Well it should be, but many developers struggle to source the right tech to power redundant navigation systems. While there’s a wide variety of reliable, open source relative navigation solutions available today, building a universal reference frame for absolute positioning remains a challenge.
At Point One, we’re creating the most precise and accessible RTK network to power GNSS positioning around the globe. Our Atlas INS leverages this best-in-class corrections service along with tightly coupled sensor fusion to achieve centimeter-level accuracy positioning data in an absolute frame, bringing us ever-closer to the elusive universal reference required for true absolute navigation. We recently combined this tech with open source relative navigation solutions to show how easy it is to globally and locally reference LiDAR data (spoiler – it only took an afternoon of work).
We hope we’ve convinced you to abandon your previously held opinions on absolute and relative navigation. To get started optimizing your own autonomous product’s navigation system, check out our dev kits.
