Projects using artificial intelligence – Project Maven – Automatic target detection from unmanned aerial reconnaissance imagery

1. Introduction.

This article introduces Project Maven, a new AI (Artificial Intelligence) project in the USA, which uses AI to analyze videos and images taken from unmanned aerial vehicles (UAVs) to identify and locate objects such as buildings and people. The word “maven” here means “expert, master, connoisseur”. In other words, Project Maven is a project that aims to use AI to interpret images from aircraft, which were previously interpreted visually by experts, in a specialist-like manner.

2. Before Project Maven

It is well known that the US military has carried out operations using UAVs in the past. In particular, several books have detailed cases of attacking the terrorist using UAVs in Afghanistan and other parts of the Middle East [1][2]. In these cases, the National Geospatial Intelligence Agency (NGA) personnel were responsible for interpreting the attack targets from the images taken by the UAV and determining their location, while Central Intelligence Agency (CIA) personnel and soldiers were responsible for carrying out the attack. The literature [1][2] describes in detail how the two work together, sometimes in the same container at the local base, to interpret targets and carry out attacks.

3. The genesis of Project Maven

Project Maven became publicly known in April 2017 [4]. It was subsequently described in some detail by Marine Colonel Drew Cukor at the Defense One Tech Summit in July 2017 [4]. According to him, by the end of 2017, the MoD will have advanced computer algorithms on government platforms that will be able to extract targets from mass video and still images. Cukor said in his remarks.“People and computers will work symbiotically to increase the ability of weapon systems to detect objects,” Cukor added. “Eventually we hope that one analyst will be able to do twice as much work, potentially three times as much, as they’re doing now. That’s our goal.”

Currently, the current target for extraction is the extraction of 38 object types needed to fight the Islamic State in Iraq and Syria. At this point, however, no actual examples were yet given.

Project Maven in concrete form finally came into the public domain in December 2017 in an article in Defense One [5]. In this article, the author wrote: ‘The Pentagon’s New Artificial Intelligence Is Already Hunting Terrorists. After less than eight months of development, the algorithms are helping intel analysts exploit UAV video over the battlefield’, the catchy title of which first caught the eye. The compendium of this article is as follows.

Earlier this month, computer-assisted analysts in identifying objects from video of UAVs on the battlefield at a location in the Middle East. Within a few days of the test run, the computer identified 60% of the types of objects – people, vehicles, and buildings – and over the next week or so, after repeated software updates on the site, the accuracy improved to about 80%. Next month, when the software and hardware are updated, the accuracy will be even better, Air Force Lt. Gen. John N.T. “Jack” Shanahan said. The project, called Project Maven, extracts objects from video taken from the small ScanEagle to help the Special Operations Command analysis support. When deployed, it will fly against a different environment than the one in which the computers were trained. In this case, it will, of course, be operational, but it will be important to optimize and refine the algorithms to suit the local conditions over the first five days of live deployment. The team is combining the Maven algorithm with a system called Minotaur, a Navy and Marine Corps “correlation and geo-registration application”in which Maven classifies objects on a video screen and tracks them. Minotaur is then used to establish the coordinates and display the object’s location on a map.

4. Latest status of Project Maven

Project Maven continues to expand its project scope despite various controversies, and more certainly its sophistication and deployment in actual warfare has become apparent.

First, at the 2022 GEOINT Symposium, NGA Secretary Robert Sharp announced that operational control of Project Mavens’ GEOINT AI services and capabilities had been transferred to the NGA from the Office of Undersecretary of Defense for Intelligence and Security [7]. According to the NGA, Project Maven’s increasing sophistication and real-world deployment are now underway and deployed in Ukraine by a “military partner”[8].

The NGA has also begun seeking private-sector help to build Project Maven from April 2023 onwards. As part of Project Maven, the MoD announced the non-competitive purchase of its proprietary software, the Maven Smart System, from the data fusion company Palantir,[9] starting with a further major contract of $480 million in May 2024, and in Project Maven, the participation of the private sector has become a must.

How usable is Project Maven at present? NGA’s Mark Munsell outlined the expected performance at the GEOINT Symposium in 2023. “I have analysts of course that would expect things to be 100 percent accurate.” He said “I have combatant commanders that will expect the object identification to be 100 percent and the geolocation accuracy to be two meters or less so they can deliver precision guided munitions. Those are the expectations. Are we there? We’re not there. Will we get there? Absolutely. It’s inevitable.” [11]

5. Conclusion.

It is unclear whether UAV attack autonomy, which automatically launches attacks against targets detected by Project Maven, has been adopted in current US military operations, as it has not been publicly announced. However, at least the ‘collection’ part of operational-level intelligence has been automated and real-time using AI. Whether the type and location of targets extracted by Project Maven are then ‘distributed’ to local commanders or soldiers, who then decide on the GO/NOGO of the attack based on human judgment, or whether the attack is launched automatically as it is, is another matter.


[1] Brett Velicovich, Christopher S Stewart

Drone Warrior: An Elite Soldier’s Inside Account of the Hunt for America’s Most Dangerous Enemies – June 27, 2017

[2] Richard Whittle

Predator: The Secret Origins of the Drone Revolution – September 16, 2014

[4] Project Maven to Deploy Computer Algorithms to War Zone by Year’s End (2017)

https://www.defense.gov/News/News-Stories/Article/Article/1254719/project-maven-to-deploy-computer-algorithms-to-war-zone-by-years-end

[5] The Pentagon’s New Artificial Intelligence Is Already Hunting Terrorists (2017)

https://www.defenseone.com/technology/2017/12/pentagons-new-artificial-intelligence-already-hunting-terrorists/144742

[7] Pentagon’s flagship AI effort, Project Maven, moves to NGA, Breaking DEFENSE, April 27, 2022

https://breakingdefense.com/2022/04/pentagons-flagship-ai-effort-project-maven-moves-to-nga

[8] What We Know About Project Maven, Reapers, and Ukraine, Tech Inquiry, 2023-03-15 https://techinquiry.org/?article=maven-reapers-ukraine

[9] Pentagon certified Palantir as only supplier for artificial intelligence targeting tool known as “Maven Smart System” by JACK POULSON All-Source Intelligence Fusion, 2023/05/14 https://jackpoulson.substack.com/p/pentagon-certified-palantir-as-only

[10] Palantir lands $480M Army contract for Maven artificial intelligence tech

https://defensescoop.com/2024/05/29/palantir-480-million-army-contract-maven-smart-system-artificial-intelligence/

[11] Despite Advances in AI Technology, it’s Still Not ‘Good Enough,’ NGA Official Says, ViaSatellite, June 1, 2023  https://www.satellitetoday.com/government-military/2023/06/01/despite-advances-in-ai-technology-its-still-not-good-enough-nga-official-says/ Each URL was confirmed on 5 August 2024 for all of them.