Part One - Requirements and Alternatives
As the autonomous outdoor robotics market accelerates toward $14.2 billion by 2032, industries face a critical inflection point. This explosive growth isn't just about technological advancement, it's a direct response to severe labor shortages, skyrocketing operational costs, and unprecedented demands for 24/7 productivity. Whether fully autonomous or human operated, there are many outdoor applications where demand for automation is growing, including agriculture, construction, defense, logistics, delivery and mining. Location aware perception is the capability of the machine to know where it is and understand its surroundings. It is a critical technology for machines to perform complex tasks with higher efficiency and accuracy, thereby critical for adoption of automation across various industries.
Choosing the right localization and perception technology for an outdoor mobile machine is not trivial. This series of blog posts will analyze the different perception technologies to consider. Part 1 will discuss the requirements, the available technologies and the criteria for evaluation.
The Complex World of Outdoor Localization
For an outdoor autonomous or semi-autonomous machine to operate effectively and be widely adopted, it needs to know where it is, where to go, and understand its surroundings on the go. In other words, it needs Location Aware Perception. Such a solution is needed to enable the mobile machine to:
Operate anywhere & anytime including in scenarios where GPS is unavailable, lighting is poor, in varying weather conditions, and when the machine needs to partially drive indoors.
Be easily deployable in large, unfamiliar areas with dynamic obstacles like humans, vehicles, and animals.
Localization is a critical basic layer for perception. Without reliable localization, the machine (or the human operator) cannot know where it is and where it needs to go. It has to work reliably, everywhere, all the time.
Consider a simple yet common scenario: If a mobile machine operated by a human loses GPS position for a while, it’s not the end of the world. The operator will handle it. But if an autonomous agriculture robot loses its positioning, even for just a few meters, it’s a showstopper, literally. It must stop and wait for assistance. This is not a rare case, it happens frequently in outdoor environments like near tall buildings, under trees, next to a wall, or within indoor transitions. Even this seemingly minor disruption requires maintaining a costly on-call human operators that combined with the loss in productivity due to down time, often defeats the core purpose of automation. In many applications the margins are so tight that such reliability issues can make the difference between a viable automation project and one that fails to deliver sufficient value to justify its implementation.
Mapping the Landscape of Localization Techniques
When examining outdoor localization technologies, we will use three evaluation criteria:
Coverage & Availability: The types of environments and scenarios in which the localization solution works well or fails. For example, GNSS (GPS) based solutions work well in open sky environment with line of sight to satellites but fail when it is blocked or when there is multipath or RF interference. 3D Lidar works well when there are static 3D objects around but fails where there are none within its range.
Reliability & Error Behavior: The reliability of the localization information reported and the ability of the localization solution to identify and correct errors. Error behaviors can be divided into two main categories:
Drifting Behavior: Technologies that accumulate error without any mechanism to reduce or recover from it. For example, inertial or visual odometry cannot recover its drift.
Bounding Behavior: Technologies that have an internal ability to reduce error and statistically bound it. Only technologies with error-bounding behavior can recover from failures without external user intervention. GPS based triangulation is a good example of error-bound technology.
Dependency on Infrastructure: Some technologies rely on installation of dedicated infrastructure in the environment to provide localization information. Such installations may not be viable in large, inaccessible, or unknown areas and often require investment and resources for installation and maintenance. Infrastructure could be global (like satellites, or Cellular antennas) or local (like RF beacons, reflectors, visual markers etc.).
A Deep Dive into Localization Technologies
Below is a high-level survey of common outdoor localization techniques and technologies:
GNSS RTK: Provides very reliable and highly accurate (1-2cm) localization information with error-bounding behavior but depends on the machine having line-of-sight to satellites. It further relies on differential GNSS that requires connectivity between the robot and a reference station on the ground. Solutions based on this technology struggle in areas with insufficient coverage (e.g. trees, buildings) or signal multi-path (e.g. urban environments) and do not work in indoor or semi-indoor scenarios (e.g. underground parking, mining sites).
Inertial Odometry: Extends localization information by integrating signals relative to a previously known location using inertial information (IMU). This technology drifts significantly over time and provides limited coverage and reliability in conditions like skidding or slippery surfaces.
Visual Odometry: Extends localization information by integrating visual signals relative to a previously known location using computer vision algorithms. This technology drifts significantly over time and provides limited reliability in conditions with no clear visual marks or challenging travel conditions.
Visual SLAM: Provides full localization information (6DoF) using computer vision algorithms for simultaneous localization and mapping based on cameras. Visual SLAM is not error-bounding by itself (drift is measured by % of the traveled distance) but when combined with loop-closures (when the robot returns to the same point), solutions based on this technology are error-bounding and provide good coverage where there are distinguishable visual features and sufficient lighting.
Local RF/UWB: Provides error-bounding localization information but depends on fixed physical installations. Solutions based on this technology offer good coverage when line of sight to the infrastructure can be guaranteed but suffer from coverage blind spots and reliability challenges in environments with many metal installations.
Laser meters: Solutions like Leica, Trimble, Faro, Topcon etc., that use an optical system with a static base (usually on a tripod) and direct optical distance and angular measurement to a target reflector. It can be very accurate but requires a direct line of sight to the static base.
3D/2D Laser scanners: 2D LiDAR offers limited coverage due to its 2D view and struggles with uneven surfaces and dynamic environments. 3D LiDAR provides comprehensive spatial awareness and better error-bounding behavior but is expensive and complex to implement. Both 2D and 3D Lidar solutions are sensitive to dust and weather conditions and are limited by the Lidar detection range.
Vision/Lidar based anchoring: Provides positioning information by comparing input from sensors to a pre-existing map of a given area. This technology relocalizes and does not provide positioning information by itself and is therefore error-bounded when there is a match between the sensor input and the pre-stored map but might lose track when there is no match. This technology depends on the ability to pre-map and for the environment to remain like the map. When the environment is too dynamic, this technology is likely to fail.
Multi-Modality - Key for Coverage & Reliability
Experience with various machines across different outdoor use cases has shown that having at least one error-bounding modality active at any given moment is essential for achieving coverage, availability, and reliability. Moreover, combining multi-modalities, by using different types of sensors, improves overall robustness. If done correctly multi-modality can guarantee that failure of one sensor will not fail the system.
RGo Perception Engine combines Visual SLAM with GNSS-RTK, inertial, and wheel odometry within an advanced sensor fusion engine to provide full localization information (6DoF) to meet the highest performance level. The addition of the learning vision system used in the Perception Engine adds visual anchoring to further improve the performance, without requiring any infrastructure or installation.
GNSS-RTK: provides coverage in open areas.
Visual SLAM: complements GNSS-RTK by providing coverage in non-open areas.
Inertial & Wheel Odometry: provide reliability and temporal coverage in GNSS-RTK denied areas with insufficient visual information.
A powerful multi-modality sensor fusion software further improves reliability and performance of the Perception Engine by constantly monitoring and evaluating the confidence level of the different inputs from different modalities, enabling the localization solution to fuse the different inputs in a robust manner. For instance, when the floor is slippery, the sensor fusion engine will ignore wheel encoder data and give more weight to vision, GNSS-RTK, and inertial odometry.
The table below summarizes the main technologies available, their pros and cons with respect to coverage, reliability and dependency on installation of infrastructure.
Technology | Coverage & Availability | Reliability & Error Behavior | Dependency on Infrastructure | Notes |
GNSS RTK | Outdoor only. Limited in obstructed areas (next to trees, buildings, etc.) | Error bounding, but susceptible to multi-path issues | Requires access to RTK correction service or local antenna | Perfect for open sky applications |
Inertial Odometry | High availability. Slightly impacted by environmental conditions (temp, pressure, vibrations, etc.) | Highly drifting | None | Affordable addition to other techniques |
Visual Odometry | Requires good image and visible features in the short range | Drifting quickly over distance or at poor visibility | None | Good addition for short range tracking |
Visual SLAM | Requires some light and distinguishable features | Error bounding with loop closures | None | Complex and hard to implement but super robust when implemented well |
Local RF/UWB/ Laser meter | Good coverage with line of sight; Suffers from blind spots in complex environments | Error bounding | Requires dedicated physical installations | Good for small, open zones. RF transmission has regulatory limitations; |
3D/2D Laser scanner | Works only when there are 3D structures within range. 2D is limited to flat surfaces with clear boundaries | Unreliable in dynamic environments or between moving objects | None (as long there are 3D structures within range) | Expensive and has limited resolution, range and field of view |
Multi-modality solution based on: Visual SLAM, Inertial + Visual Odometry and GNSS RTK | Very Good, everywhere | Error bounding | None Requires access to RTK correction service or local antenna if using RTK | Provide best coverage, highest availability and reliability for multiple use cases at reasonable cost |
Key Takeaways
Building autonomous or semi-autonomous machines for outdoor use is a complex challenge that requires advanced technology and expertise. From agriculture and construction to logistics and defense, these machines must operate reliably in diverse and often unpredictable environments. Key to their success is a robust perception system that ensures accurate localization, even when GNSS GPS signals are weak or unavailable.
A range of technologies are available to build a localization solution, each with its limitations and failure cases. Combining multiple technologies and modalities is essential to achieve comprehensive coverage and reliability.
RGo’s artificial perception software platform leverages advanced AI, sensor fusion, and learning algorithms to provide a solution that works anywhere, anytime. RGo’s platform was tested in a range of outdoor environments and offers breakthrough accuracy, adaptability to environmental changes and cost-effectiveness, making it the ideal choice for companies looking to stay ahead of competition.
Want to learn more about how RGo Perception Engine can help your outdoor machine perform better? Contact us today!
Comments