Unmanned aerial vehicles (UAVs) are quickly emerging as a viable, low-cost technology for use in various indoor applications (see Figure 1). These services and applications often demand precision UAV autonomy that requires highly refined control dynamics and strategic mission planning, both of which are contingent upon accurate determination and localization of the deployed UAV.
Figure 1: Potential applications of indoor localization of UAVs
Although outdoor positioning of UAVs has greatly benefited from the advancement of a global positioning system (GPS) coupled with onboard inertial navigation system (INS) measurements, GPS-based localization remains impractical in closed environments due to the high attenuation of GPS signals across enclosing materials, resulting in non-negligible positioning errors. This is because functioning GPS triangulation assumes a direct line-of-sight propagation of signal between the satellite and the GPS module (receivers) to extract positioning information, based on the velocity and the travel time of transmitted signals.
Any physical barriers, such as materials or human presence, can drastically affect the signal travel time, potentially skewing the results. Therefore, an alternative method has to be developed for indoor positioning for UAV applications with a design emphasis on resolution, mobility and infrastructure minimalism. See Figure 2 for existing solutions and descriptions.
Figure 2: Existing indoor positioning solutions
Design considerations for indoor localization of UAVs
If we were to informally categorize the different scales of control within an enclosed space, it can be broadly generalized into two: micro-control and macro-control. In a closed environment, macro-control in our definition refers to the control of UAVs from one 3D (x,y,z) point to another (i.e., its maneuverability across regions or chunks of space within the entire environment). If so, micro-control focuses more on the control around a point space of reference within a chunk of space and its maneuverability in the space around that point.
Most aforementioned indoor positioning systems are able to provide resolution up to a few meters, which allows dynamic macro-control in the closed environment. However, in most instances, UAV applications require both precision micro- and macro-control; as such, a number of additional considerations must be factored in while selecting a solution to achieve indoor location of UAVs:
Resolution – To achieve precision in both micro- and macro-control of UAVs, both spatial and temporal resolution of the localization or tracking system have to be considered. Spatial resolution of the system determines how finely we can track UAV movements around a point space. To achieve basic capabilities such as obstacle avoidance, a spatial resolution of less than one meter error is required.
Temporal resolution refers to how accurately the position estimation computed by the back-end tracking system is reflective of the UAV’s actual position at the time; in other words, the time it takes for a change in UAV position to be detected by the visioning system. This is equally important considering much of the real-time control will be highly reliant on assuming the tracked position closely aligns with the UAV’s current position.
Accuracy – This refers to the mean baseline estimation error in the localization system. Depending on the error margin, the envelopes of the micro-control space have to be adjusted to provide a degree of confidence, thus limiting the boundaries of micro-control capabilities.
UAV autonomy/infrastructure independence – The UAV platform should ideally be autonomous and adaptable to unfamiliar environments for varying applications, suggesting that much of the localization and processing should be onboard rather than having the need to install additional dedicated hardware onto the existing infrastructure.
With these design considerations in mind, the trade-offs between using various indoor positioning systems for UAV localization quickly become apparent (see Figure 2). To achieve both effective micro- and macro-control, only using a network of cameras offers the adequate spatial resolution (cm) but it is not cost effective for scaling to larger or other indoor environments for two reasons. One, there are limitations to the range and field of view of each camera, so many more cameras are needed if the indoor space increases; two, each additional indoor environment would require additional cameras, and would be unable to utilize existing infrastructures.
Other solutions, such as the differential WLAN access points, effectively exploit the infrastructure in place and are able to attenuate environmental fluctuations, making it an easier and more reliable solution to adopt. However, it does not offer enough spatial resolution to achieve micro-control required by most UAV applications.
Solution: Move toward onboard positioning solutions
In order to design a more scalable, robust and autonomous UAV localization solution, it seems reasonable to migrate slowly away from off-board indoor positioning systems and push more sensors, intelligence and perception onboard. We can definitely use off-board localization solutions, such as the WLAN access points, for macro-control considering its ease of integration across various environments. However, micro-control eventually has to migrate completely to onboard computation for vision and probabilistic methods, such as simultaneous localization and mapping algorithms (SLAM), to achieve full autonomy in an unknown indoor environment without a priori knowledge about the infrastructure.