Drone AI and the benefits to mission autonomy 

Non-kinetic UAV operations could be aided through the greater use of AI in data processing and long-endurance missions. Gerrard Cowan reports.

Artificial intelligence (AI) could bring a range of advantages for uncrewed aerial vehicles (UAVs), notably when it comes to processing data and boosting autonomy. The technology is being pursued through a range of projects, with developers seeking to overcome challenges to increase the sophistication and widen its availability.  

UAVs can produce far more sensory material than humans can easily process, noted Greg Sanders, deputy director and fellow in the Defense-Industrial Initiatives Group at the Center for Strategic and International Studies, a US-based think tank. Lower latency could be one advantage of onboard AI, he said, because it means processing can be done in fewer steps. However, there are two wider benefits. 

First, AI cuts the amount of data that needs to be transmitted mid-mission. This “reduces observability and may help prioritise information to servicemembers in support of flexible kill chains”, Sanders said. Second, “autonomy becomes especially important if countermeasures cut off remote control”.  

There are a range of ongoing efforts in the area. For example, Lockheed Martin is partnering with software specialist Red Hat to run AI models at the edge on the Stalker UAV. This is powered by Red Hat Device Edge, a platform that combines lightweight Kubernetes using MicroShift with Red Hat Enterprise Linux.  

“Data processing is a popular use case for onboard AI models,” said Tony James, chief architect for science and space at Red Hat. “Being able to analyse data using onboard compute resources helps save time by reducing the need to send and receive large datasets.” 

The process for deploying AI models depends on the software platform that is used, said James. “Containers are a popular method for this use case because they provide a consistent experience from development to edge deployment.” 

Sentient provides AI software solution for a range of projects, notably its ViDAR optical radar, designed to provide autonomous AI search and detection capabilities. It has been deployed on the Insitu ScanEagle, while the company is currently working on installing the system on Stalker UAVs from Lockheed Martin, among other focuses in the defence space.  

The biggest challenge for UAVs today is the datalink back down, said Mark Palmer, acting CEO at Sentient Vision. “Most operations of UAVs today rely on the data being brought back to a central point and processed at that point, whether that be video feed or something else,” Palmer said. “Most UAVs also rely on a constant datalink to control them, and that's been shown to be quite an issue as well.” 

The ability to process data onboard expands the amount of data a UAV can effectively gather and process because the bandwidth limitations are reduced. Another alternative is capturing data, landing and then removing the data and having it processed, but this is slow. 

“Now it's really the next generation of UAV operations, where UAVs can go out there, they can basically handle incredible amounts of data, because they process it on board, and they send back just the processed data to whoever wants it,” Palmer said. “That's a considerable reduction in the data that’s sent back and therefore a considerable increase in the amount of physical data they can process.” 

// A number of international navies and coast guards have operated ViDAR-equipped ScanEagle UAVs to provide autonomous search and tracking capabilities. Credit: Insitu

A ViDAR-equipped ScanEagle UAV embarked onboard a US Navy warship 

Continuing, Palmer said intelligence, surveillance and reconnaissance (ISR) are the key missions that benefit from onboard AI today, because it effectively increases the amount of land or ocean the system can process when it comes to data. Additionally, if the system is performing a covert operation, processing the data onboard and only transmitting it in limited bursts means it is less likely to be detected.  

Sanders also pointed to the value of commercial advances in AI when it comes to ISR, though he noted that “human-in-the-loop” demands might limit some potential uses of AI. On the whole, contested operating environments could be where AI proves most valuable, though it will be vulnerable to AI-specific countermeasures.  

The size of the system is also directly relevant to its capacity to support AI. As Sanders noted, “larger systems inherently have more room for computing hardware and the power generation to support it”. 

However, Sanders said that interoperability and modular open systems could provide benefits, making it easier to deploy upgraded software and sometimes hardware components for a range of different systems.  

Meanwhile, Palmer said Sentient aims to extend the capacity of its AI-powered optical radar to higher altitudes, covering extremely large areas. The company is working on a new piece of equipment that will provide a 400-megapixel camera capable of analysing data over up to 100km2 areas, for example tracking every vehicle, at about 20,000-25,000ft. “That would allow a large area to be monitored from an ISR point of view at any one time,” he said.  

While it is always easier to deploy the technology onto larger UAVs, the revolution in mobile processing and cameras means they can push the technology down to very small UAVs today. On a wider level, the biggest challenge is processing power, according to Palmer. 

“Can you put enough lightweight, power-efficient processing on board? The real challenge is the amount of processing power you can get onboard, because your challenge on board any UAV is that the more power your onboard processing draws, the less power the engines have for time in the air,” Palmer stated.  

Palmer envisions a future where UAVs fly continuous automated missions independently, which could have advantages in ISR and in areas like search and rescue.  

“I think that will become a reality in the next few years. And we'll see more and more of that. We'll see drones that are on continuous patrols. and we'll see the ability for the drone to map huge areas or patrol huge areas that we never really see today.” 

James said that with advances in computing happening all the time, it will be interesting to see what the future brings. “With technologies like ChatGPT from OpenAI there may be a future where fully autonomous AI models are able to run at the edge to make decisions in real time,” he said. 

We want to be faster, inside any threat, or enemy decision cycle, the notion in being far more agile in that decision dominance

// Main image: The ability for UAVs to operate more autonomously when conducting missions such as mapping would bring a number of benefits, with bandwidth a limiting resource. Credit: Sentient Vision