The use of commercial drones and the value of their markets are both continually on the rise, with a number of different sources forecasting the global drone industry to surpass $50 billion this decade. In order to continue the increase of these levels of innovation and scale, operating models need to move beyond the standard paradigm of a single drone operated by a single pilot within VLOS (visual line of sight).
One way to drive this innovation is the use of artificial intelligence (AI). Once considered the domain of science fiction, AI has benefited from recent improvements in processing technology and miniaturization and is now ideal for providing drones with enhanced autonomy as well as advanced intelligent data analysis.
The application of AI can allow drones to achieve results quicker and more accurately than pilots and operators are able to and pave the way for taking humans out of the loop altogether. Besides the increase in efficiency and cost-effectiveness that this will bring, AI could also enhance the safety of unmanned aviation by removing human error and operator fatigue from the equation.
AI vs machine learning vs computer vision
Before we explore the uses of AI for commercial drones further, it is worth looking at some of the terminologies that are often used adjacent to AI, and how these concepts differ technically.
Artificial intelligence is a highly broad set of technologies that allow computers and robotics to simulate aspects of human intelligence such as learning and problem-solving. AI encompasses a variety of different approaches and algorithms that allow a computer to utilize information and make intelligent decisions.
Machine learning (ML) is a particular subset of AI and one of its most popular applications. It uses algorithms such as neural networks to allow a computer to learn from provided information and the environment around it and solve problems without the need for direct instruction. Unlike traditional software algorithms, machine learning algorithms are usually designed to improve over time with exposure to new data.
Going down another level, computer vision (CV) is an application of machine learning. It allows computer systems to analyze image and video data, spot patterns and extract information, and make decisions based on these results. Computer vision-based object recognition, identification, and detection are highly useful applications for a variety of autonomous drone operations, including navigation and detect-and-avoid. These tasks can also be used for enhanced data processing in a number of specific market segments such as utility drone inspection, precision agriculture, and photogrammetry.
AI for BVLOS and autonomous operations
In order to scale up to a new level of commercial viability, drone operations need to advance to BVLOS (beyond visual line of sight) flight. Such operations are highly regulated by most aviation authorities around the world, including the FAA and EASA, and require lengthy certification processes during which foolproof safety and reliability systems must be demonstrated.
AI and computer vision can help provide some of these essential capabilities. One of the most critical technologies for BVLOS is detect-and-avoid (DAA), which enables drones to detect obstacles and other hazards in the environment and autonomously maneuver to avoid collisions. Computer vision models can be applied to camera feeds that continuously monitor the surrounding airspace in real-time and trained to pick out other drones and manned aircraft, birds, powerlines and a variety of other potential hazards.
AI can also allow drones to make safe landings with improved accuracy, as well as emergency landings in the case of unforeseen circumstances. Combining object recognition with other inputs such as GPS receivers and wind sensors, AI-based algorithms can select a suitable landing site and direct the flight computer to adjust the drone’s trajectory and speed to achieve the safest possible touchdown.
In addition to landing, other precision maneuvers may also benefit from the application of AI drone technology. Similar object detection and precise positioning algorithms could be applied to drone delivery, which may need to take into account the environment around the selected drop zone and ensure that the cargo is released without any risk to people, property or the package itself.
A solution for drones on a lower SWaP budget
While some AI applications only require post-mission processing, others, such as object tracking, need data to be analyzed during flight in or close to real-time. Such capabilities come at a cost, as onboard embedded systems that can handle this complex processing are typically highly compute-intensive and can vastly increase the SWaP (size, weight, and power) requirements of the drone.
One way to allow drones with lower SWaP budgets to still take advantage of AI is to offload the processing to cloud servers and edge computing services. Many regions covered by 4G LTE and 5G cellular networks will provide the necessary bandwidth and latency for this approach to be successful.
If you are designing a drone system that uses AI and computer vision to facilitate BVLOS operations, your cellular communications link needs to be rock-solid in order to persuade regulators that your platform is safe to fly. This is where Elsight’s Halo comes in.
Halo is a compact and lightweight connectivity solution that provides maximum connection confidence for BVLOS UAS drones. Using advanced bonding technology, it can aggregate bandwidth from up to four separate cellular connections into one secure datalink. It also provides automatic traffic balancing that adapts to the dynamic needs of any drone mission and allows the communications system to switch to a backup link in case of network coverage issues, providing maximum uptime and reliability for your aircraft.
How can autonomous drone operations be made safer?
Autonomous drones can be equipped with a number of technologies that will enhance the safety of their operation. These include detect-and avoid (DAA) systems, parachutes, and reliable communications links with built-in redundancy.
What do drones use artificial intelligence for?
Drones may use artificial intelligence technologies for a range of different problem-solving and decision-making applications. These include computer vision and machine learning for collision and hazard detection, precision landing, and autonomous inspection and monitoring.
Do drones have AI onboard?
Many drones have onboard computing capabilities that can handle the processing of artificial intelligence algorithms and data. As this processing can require a significant amount of power, smaller drones with lower SWaP budgets may not be able to incorporate onboard computing. As an alternative, they may be able to send data to remote cloud servers to be processed.
To find out how Halo can be an asset to your AI-powered or data-intensive commercial drone platform, please get in touch to find out more.