Home : News : Display
March 6, 2026

Successful Large-Scale Combat Operations Require Artificially Intelligent Breaching Munitions

Michael P. Carvelli

ABSTRACT: The US Army must invest in artificial intelligence–enabled breaching munitions to succeed in large-scale combat operations. This article combines existing technologies to propose a new capability the Army does not currently possess. It lays out the current state of breaching materiel, identifies additional technology available, and proposes combining multiple pieces of existing technology to create improved breaching munitions for the Army’s use. This new capability will require practitioners and policymakers to enable the creation of artificial intelligence systems through acquisition and tactical experimentation.

Keywords: artificial intelligence (AI), countermine operations, large-scale combat operations (LSCO), breaching tactics, machine learning (ML), Russia-Ukraine War, Nagorno-Karabakh

 

The US Army must push forward the concept of artificially intelligent breaching munitions (AIBMs) to win against peer and near-peer threats in large-scale combat operations (LSCO). These munitions do not exist yet, but technology is reaching the point where several technologies can be mated to create AIBMs. To ensure the United States maintains a position of relative advantage over potential adversaries, the Army must accelerate the production of AIBMs.

This article proposes a concept that places breaching at the nexus of artificial intelligence (AI), unmanned aircraft system (UAS)–based sensors, ground robotics, and human supervision (referred to as human in the loop). Imagine a swarm of UASs flying in disparate paths over enemy-held terrain. Armed with sensors capable of detecting myriad obstacles, the UAS swarm would transmit or record data on enemy positions, and artificial intelligence (AI) would then sift through the data, using image recognition to identify and georeference obstacles. Once AI identifies the obstacles, the friendly commander would develop a scheme of engineer effort whereby swarms of AIBMs would detonate breach lanes, at scale, to support friendly maneuvers. Throughout the process, soldiers would validate and redirect the process as necessary without ever being physically present.

Creating AIBMs would make breaching safer, faster, and more scalable—safer because remote validation would remove soldiers from the breach, one of the most dangerous points on the battlefield; faster because AI can perform the same tasks as humans at a faster rate and with more predictable results; and scalable because, from squads to battalions, engineers could use AIBMs on a range of obstacles, from a single dismounted, mine-wired obstacle to multiple complex, mounted obstacles.

Observations

The Russia-Ukraine War highlights the urgent capability gap in this technology. In Ukraine, both sides are deploying millions of low-cost drones that play the roles of scouts and weapons in combat. The use of drones, also known as UASs, has allowed both sides to inflict personnel and equipment losses and conduct surveillance and reconnaissance. Small drones are already a proven wartime technology. The United States can expand on the current capabilities of drones but must do so quickly.1

As observed in Ukraine, Russian doctrine emphasizes the attrition of enemy forces, using land mines as a vital component of the process. As of 2019, Russia possessed the world’s largest stockpile of antipersonnel (AP) mines, with estimates placing the number of mines around 27 million. Russia has deployed 13 variants of AP mines and another 13 variants of anti-tank mines, some not previously seen in conflict, all over Ukraine. Russia has also used its stockpile of AP and anti-tank mines in Russian defenses, and Ukraine has suffered terrible losses in its attacks on Russian fortified positions. Some of this attrition can be attributed to Russia’s magazine depth—specifically its stockpile of mines. Mines will persist in warfare.2

After Russia invaded Ukraine a second time in 2022, Russia created miles of defensive lines. For example, the Surovikin line is a complex system of Russian defensive fortifications and obstacles that includes anti-vehicle ditches, concrete pyramids (known as “dragon’s teeth”), and trenches. Rather than being a single, contiguous line, this line includes multiple features that connect and intertwine in three layers. The first layer, an anti-vehicle ditch, prevents tanks, infantry fighting vehicles, and other heavy armor from advancing. The second layer, a row of dragon’s teeth (concrete spikes) is designed to stop vehicles. The third layer consists of trenches manned with Russian soldiers who can observe the ditches and dragon’s teeth from their fighting positions. The Surovikin line presents a formidable obstacle traditional breaching techniques struggle to penetrate.3

After two years consolidating its gains, Russia continues to reinforce these defensive lines with complex counter-mobility obstacles. Ukraine’s army is innovating as rapidly as possible with the tools it has on hand. Progress is slow, however, as Ukraine is wed to existing equipment and tactics supplied by Western countries. The US Army needs to look beyond the current situation and shape the desired future state through acquisition and innovation.4

Current State

A revolution in military affairs (RMA) may be imminent due to the proliferation of UASs. One could define an RMA as a profound transformation in war fighting. The proliferation of UASs has made precision-guided munitions available to countries previously unable to afford them. No longer do countries need an acquisition ecosystem to research, develop, test, and evaluate precision-guided munitions. Ukrainian citizens can mass-produce UASs, mate them with an explosive, and destroy Russian equipment for hundreds of dollars. This RMA is evident beyond the Russia-Ukraine War.5

For example, in the second Nagorno-Karabakh conflict (2020), Armenia and Azerbaijan employed a mix of domestically produced drone variants and systems from other countries. This conflict highlighted the weaknesses of armored platforms when faced with explosively armed UASs (referred to as loitering munitions)—an improvised loitering munition costing a few hundred dollars could defeat a million-dollar tank in a single attack. By no means are tanks obsolete; they have their place in battle. The proliferation of loitering munitions, however, has changed the conduct of modern war.6

The Ukrainian army has shown its ability to incorporate emerging technologies rapidly during tactical operations. Ukraine recently executed an attack on Russian positions using only uncrewed ground robots and UASs. Allegedly, this attack represented the first instance of an uncrewed battle in the Russia-Ukraine War. Ukraine has also attacked Russian forces with UASs that have automatic target recognition enabled. These emerging technologies indicate an RMA has occurred, and the Army must adapt.7

Today’s RMA recalls the eclipse of medieval knights in the fifteenth and sixteenth centuries, when the use of gunpowder in cannons made these heavily clad warriors obsolete. The development and employment of guided anti-tank munitions did not make tanks obsolete; rather, these munitions displayed the vulnerability of the tank, forcing tactics and protection systems to change.8 To enable the advancement of maneuver formations to close with and destroy the enemy, Washington must advance the use of AIBMs, lest the Army be forced to change tactics or find its equipment obsolete during a conflict. The Army must break its cycle of making only incremental improvements with deliberate, long-range innovation of AIBMs.

The Army depends almost entirely on manned mobility capabilities, including assault breacher vehicles (ABVs), towed mine-clearing line charge (MICLIC) trailers, and dismounted squads using various hand tools (such as grappling hooks and wire cutters). These systems have existed for decades and have their place in the tool kit. The Army posses at least one robotic breaching system—the M160 MV4 DOK-ING, a robotically operated tracked mine flail—and is working toward creating robotically operated dozers and ABVs, among other systems. For example, the 20th Engineer Brigade has experimented through the Sandhills Future Breaching Experiment to deliver robotic, autonomized breaching solutions for LSCOs. The brigade uses existing LIDAR (Light Detection and Ranging) technology and other methods to increase the proficiency of these unmanned systems, mainly through heavy human integration. Unfortunately, these systems are not AI enabled. While the Army has started pursuing the use of machine learning (ML) for breaching, the technology is nascent and not yet fielded. The Army must move faster.9

Aside from mostly crewed systems, the Army continues to rely on large, expensive breaching implements it does not possess in significant numbers. One example is the M1150 ABV: a breaching vehicle created from the M1 Abrams chassis that can reduce, proof, and mark a mounted breach lane. Unfortunately, each combat engineer company (armored) has between two and four ABVs, whose operational readiness rates float between 50 and 70 percent. Additionally, ABVs suffer from a lack of repair parts and logistical expertise due to the low inventory of these vehicles. This scarcity means if one company were assigned to a single combined-arms breach, only one mounted lane could be constructed, with one ABV serving as the primary system and the other serving as the backup. Massive, heavy, expensive, and few in numbers, ABVs are only one tool in the breaching tool kit.

The Army appears wed to the idea that it must haul a rocket-launched MICLIC to the point of the breach. This tactic is one way the ABV enables the breach. In a light formation, a vehicle must tow a MICLIC trailer to the lead edge of the minefield, where engineers can employ the system. Since the MICLIC trailer is inherently less protected than an ABV, the trailer is less survivable. Unfortunately, the Army currently operates under the assumption that line charges are the only method of clearing obstacles containing mines. The time to create more options has arrived.

In any situation where a vehicle is towing or hauling a MICLIC into position, a well-trained anti-tank, guided-munition gunner can immobilize, if not destroy, the vehicle, leaving the MICLIC useless. These large systems should be attritted quickly in LSCOs. To quote General James E. Rainey, then–commanding general of US Army Futures Command, “We still have soldiers driving up [with MICLICs]—we’ve lost our mind.” To be sure, removing the soldier from the ABV or towing the MICLIC trailer makes breaching safer. While this step does not make breaching faster or more scalable, breaching robots are only on one side of the spectrum.10

The Army needs to explore the AIBM concept, using swarm tactics involving small, explosively armed robots on the other side of the spectrum. With limited resources and enormous acquisition costs for large breaching systems, small systems could bring scalability, make breaching faster at potentially reduced costs. The infusion of AI could remove soldiers from the breach, reduce the need for human scouts, and increase the speed of the decision-making cycle.

Available Technologies

Artificial Intelligence: An explanation of what AI offers can be helpful. At the core of AI lies ML, which uses computing power to learn laws from historical datasets through related algorithms, make predictions or judgments about new sample data, and learn like a human being. Inside the ML category is deep learning (DL), which imitates the mechanism of the human brain to interpret data such as text, sound, and imagery. Multiple DL models exist, but this article focuses on the outcome of image recognition.11

Image recognition and object detection involve detecting and locating predetermined objects in dynamic and complex environments. Creation of a DL model requires four main steps. First, prepare the training data, consisting of a collection of images sorted into specific categories. Next, use a pretrained model or create one using existing images that humans have labeled. Third, present the test data and iterate. Iteration allows the model to learn the most important features relative to the images, increasing the model’s accuracy and precision. Last, test the data on images that were not part of the training set. As humans train the algorithm, the algorithm increases its accuracy and precision. After humans train the model, they test it with new images kept apart from the training set, allowing the algorithm to prove its worth.12

Developing the DL model is the first major step in creating AIBMs. To focus on mines, the Army would need to feed a large training set of images into an algorithm. Fortunately, a team from Columbia University and the State University of New York at Binghamton has tested this technology by training an algorithm to detect the Russian PFM-1 AP mine with 91.8 percent accuracy in sand, grass, and rubble environments. This controlled study proves image-recognition AI can be trained on Russian mines. While great news, the technology has inherent limitations. The environments were constructed, only one type of AP mine was included, and the study occurred in a peacetime environment. Still, the study shows the Army can advance the incorporation of AI for wartime use through a similar method. Combining AI with robotics is the next step.13

To maintain a position of advantage over US adversaries, the Army should train AI to recognize mines using datasets from Ukraine and other Russian-occupied areas. Using the Russia-Ukraine War as the impetus to start this training now, the US Army can create this type of trained model.

Robotics: In conjunction with AI, the Army can use ground robotics to evolve AIBMs. Artificial intelligence increases the performance and efficiency of various other technologies and fields, including robotics. Looking solely at robotics, the US military has experienced success for decades, including QinetiQ’s TALON Medium-Sized Tactical Robot, a proven, useful technology familiar to engineers. Engineers and explosive ordnance disposal technicians have used the TALON Medium-Sized Tactical Robot as an interrogation asset in Iraq, Afghanistan, and beyond to identify threat ordnance and protect soldiers. These robots have saved countless lives by keeping soldiers away from danger. Similarly, the Army can use robotics to get soldiers out of the breach, as the Sandhills Project is proving.14

As part of a larger operation, small robots loaded with explosives could detonate over a mine or in a wire obstacle to reduce the impact of either threat. At scale, these robots could swarm a minefield or obstacle with general-purpose explosives, shaped charges, linear charges, or other capabilities. When the explosive-laden robots are directed through AI, AIBMs start to take shape. One necessary element remains—humans.

Ethical Considerations

The military profession demands that the Army incorporate ethics as it advances AI technology. Combining humans with AI could provide greater accuracy and precision. As human capacity increases due to AI, the technology could also propagate human error if humans build it with flaws or biases. Further, removing humans from the loop could increase speed and reduce human ethical oversight. With iterative improvements, AI might consider human ethical bounds as the technology progresses. If these efforts at alignment are successful, humans could move from in the loop to on the loop, meaning they would supervise the AI as it operates autonomously.

To build AI toward ethical ends, one must not merely ask how AI works. Rather, one must ask how the AI was trained. Many AI models show inherent bias after being trained on biased datasets. To incorporate ethics into the AIBM system, the Army must examine the original datasets to ensure the training datasets used to narrow the image recognition of mines are as clean as possible. If, for example, the training data included photos of mines with humans next to them, AI would plausibly direct AIBMs to destroy mines only when the latter are located near humans. While incorporating ethics when building the AIBM image-recognition model is possible, the Army must ensure this step occurs at the beginning of the process.15

Increasing Available Tools

The concept of AIBMs is better than the options available today because AIBMs would be faster, more scalable, and safer. Maneuver commanders need options, and AIBMs provide them. Swarm tactics provide these benefits and could even create decoy breaches or aid in demonstrations and feints. The Army should not replace the large systems it currently uses. Rather, the Army should add options. As the technology grows, the Army can provide AIBMs as a safer, faster, and more scalable tool.

Artificially intelligent breaching munitions would increase survivability, decrease logistics, and provide more munitions for LSCOs. Few adversary capabilities could destroy six-inch-tall robots as they maneuver to an obstacle. Such robots would have fewer parts, require few services, and be transportable on almost any platform. While ABVs provide unparalleled quality, AIBMs provide quantity. Delivering AIBMs to the last covered and concealed position would require aerial or ground delivery. Although aerial delivery requires additional work, it also provides more options, whereas ground delivery is the only option available today.

Counterargument

The emergence of UAS-based surveying has led to the new problem of analyzing datasets that are both large (in terms of area covered and file size) and prohibitively complex, requiring advances in ML to aid interpretation. Even the slightest increase in the use of AI to identify obstacles and direct AIBMs would improve the current state of the technology. Artificial intelligence continues to grow, and only through using AI can the Army understand this technology’s advances and limitations. With advances in AI will come advances in computing power.16

The concept of AIBMs is not a panacea. Some might say the way to succeed during breaching operations in LSCOs is with more human training. For example, in an article for the US Army Training and Doctrine Command G-2, Richard L. Garcia and Colin Colley note, “As minefields and other obstacles become more advanced, U.S. Army maneuver units could emphasize training on breaching deep obstacles, targeting enemy engineer assets, and the OPFOR could simulate Russian obstacle tactics.” Whereas increased training could help, the Army cannot train its way out of the current operating environment. Under conditions of constant observation and heavy indirect fires, putting soldiers in the breach will generate high casualties.17 Even robotically operated equipment (such as an ABV) will be immobilized and destroyed. Removing soldiers from the ABV provides a partial solution. Instead, massing AIBMs provides opportunity and reduces risk.18

The Army must also iterate past the pitfalls inherent in the incorporation of new technology. Separating the different causes of failure—such as sensor limitations, incorrect AI algorithms, or inadequate training data—is a prerequisite for progress. The military must resolve these situations before using AI technology in live combat. Under test conditions with a controlled dataset, AI could identify mines with more than 90 percent accuracy. Nevertheless, test conditions are different than combat. Humanitarian demining operations are executed under peaceful conditions, not among the frenzied activities of war. The Army must validate image-recognition models with high fidelity prior to using them in combat conditions.19

Finally, the Army must address the other limitations of AIBMs, including the range of the systems, the size of the munitions, those munitions’ explosive effects, and the logistical tail. After advancing the AIBM technology, the Army must assess doctrinal, organizational, and training needs, at a minimum. Providing AIBMs to maneuver commanders with confidence and at scale will require additional time and resources.

Conclusion

Enemy commanders will target engineer assets because they enable the Army’s maneuver plan, and equipment like an ABV is clearly identifiable in an armored formation. Just as the United States assesses to understand adversarial capabilities, its adversaries study US capabilities. Maneuver commanders need options, and engineers must continue to prove themselves capable. Through the concept of AIBMs, engineers can provide ethical, safe, fast, and scalable solutions to breach any adversary’s best efforts to construct obstacles. With recent conflicts highlighting advancements in technology—including AI and robotics—the Army must rapidly adopt and integrate these technologies, specifically AIBMs, to maintain America’s position of relative advantage in future conflicts.

 
 

Michael P. Carvelli
Lieutenant Colonel Michael P. Carvelli, US Army, is an assistant division operations officer in First Army Division East. He previously commanded the 1st Brigade Engineer Battalion, 410th Regiment, 4th Cavalry Multi-Functional Training Brigade, at Fort Knox, Kentucky. Carvelli holds a bachelor of science degree in civil engineering technology from the Rochester Institute of Technology, a master of science degree in operations management from the University of Arkansas, a master of science degree in civil engineering from the University of Florida, a master of arts degree in defense and strategic studies from the Naval War College, and a master of arts degree in military operations from the School of Advanced Military Studies at the US Army Command and General Staff College. He is a registered professional engineer in the state of Pennsylvania and a certified project management professional.

 
 

Endnotes

  1. 1. “Against Expensive Excellence,” The Economist, January 11–17, 2025, 23.
  2. 2. Richard L. Garcia and Colin Colley, “Russian Minefield Tactics Pose Challenge to Mobility,” Red Diamond (November 2024): 1–7; and Douglas Broom, “Two Decades Later and Illegal Landmines Are Still Stockpiled – Why?,” World Economic Forum, October 21, 2019, https://www.weforum.org/stories/2019/10/global-landmine-stockpiles/.
  3. 3. Sophia Ankel and Jake Epstein, “Ukraine Breached Russia’s Fearsome Defenses with Vehicles for the First Time, Analysts Say – A Major Milestone in Its Fightback,” Business Insider, September 21, 2023, https://www.businessinsider.com/ukraine-punches-russia-anti-tank-defenses-armored-vehicles-counteroffensive-verbov-2023-9; and Jake Epstein, “Ukraine’s Front-Line Forces Are Trying to Fight Their Way Through Russia’s Formidable Surovikin Line. Here’s What That Is,” Business Insider, September 6, 2023, https://www.businessinsider.com/ukraine-front-line-forces-fighting-through-russia-surovikin-line-2023-9.
  4. 4. Thomas Gibbons-Neff et al., “21 Miles of Obstacles,” The New York Times, June 28, 2023, https://www.nytimes.com/interactive/2023/06/28/world/europe/ukraine-counteroffensive-obstacles.html.
  5. 5. Andrew Latham, “Warfare Transformed: A Braudelian Perspective on the ‘Revolution in Military Affairs,’ ” European Journal of International Relations 8, no. 2 (June 2002): 239.
  6. 6. Eado Hecht, “Drones in the Nagorno-Karabakh War: Analyzing the Data,” Military Strategy Magazine 7, no. 4 (Winter 2022): 31–37.
  7. 7. Andrew White, “Why Ukraine’s All-Drone, Multi-Domain Attack Could Be a ‘Seminal’ Moment in Warfare,” Breaking Defense, January 24, 2025, http://breakingdefense.com/2025/01/why-ukraines-all-drone-multi-domain-attack-could-be-a-seminal-moment-in-warfare/; and David Hambling, “Ukraine Rolls Out Target-Seeking Terminator Drones,” Forbes, updated March 22, 2024, https://www.forbes.com/sites/davidhambling/2024/03/21/ukraine-rolls-out-target-seeking-terminator-drones/.
  8. 8. Sébastien Roblin, “Down 2,000 Tanks, Russia Is Using Creative New Tactics to Keep Them Alive,” Popular Mechanics, June 9, 2023, https://www.popularmechanics.com/military/weapons/a44068426/how-russia-is-keeping-tanks-alive/.
  9. 9. Justus Reed, “ERDC Supports Integration of Robotic and Autonomous Technologies,” US Army Corps of Engineers Engineer Research and Development Center, January 18, 2024, https://www.erdc.usace.army.mil/Media/News-Stories/Article/3649165/erdc-supports-integration-of-robotic-and-autonomous-technologies/; Sam Skove, “Mine-Spotting Drones and Tracked Robots: The Army’s Efforts to Breach Minefields with Tech,” Defense One, January 9, 2024, https://www.defenseone.com/technology/2024/01/mine-spotting-drones-and-tracked-robots-armys-efforts-breach-minefields-tech/393228/; and “Machine Learning (ML) for Breach Routing,” U.S. Army Small Business Innovation Research, November 16, 2021, https://armysbir.army.mil/topics/machine-learning-ml-for-breach-routing/.
  10. 10. James E. Rainey quoted in Skove, “Mine-Spotting Drones.”
  11. 11. Pin Wang et al., “Comparative Analysis of Image Classification Algorithms Based on Traditional Machine Learning and Deep Learning,” Pattern Recognition Letters 141 (January 2021): 62.
  12. 12. Jasper Baur et al., “How to Implement Drones and Machine Learning to Reduce Time, Costs, and Dangers Associated with Landmine Detection,” The Journal of Conventional Weapons Destruction 25, no. 1 (Summer 2021): 139; and “What Is Image Recognition?,” MathWorks, n.d., accessed February 5, 2025, https://www.mathworks.com/discovery/image-recognition-matlab.html.
  13. 13. Baur et al., “How to Implement Drones,” 138.
  14. 14. Gary Shapiro, Pivot or Die: How Leaders Thrive When Everything Changes (William Morrow, 2024), 135.
  15. 15. Russell Gasser, “What Can Artificial Intelligence Offer Humanitarian Mine Action?,” The Journal of Conventional Weapons Destruction 28, no. 2 (Summer 2024): 16.
  16. 16. Baur et al., “How to Implement Drones,” 138.
  17. 17. Gibbons-Neff et al., “21 Miles.”
  18. 18. Garcia and Colley, “Russian Minefield Tactics,” 4.
  19. 19. Gasser, “Humanitarian Mine Action,” 14.
 
 

Disclaimer: Articles, reviews and replies, review essays, and book reviews published in Parameters are unofficial expressions of opinion. The views and opinions expressed in Parameters are those of the authors and are not necessarily those of the Department of War, the Department of the Army the US Army War College, or any other agency of the US government. The appearance of external hyperlinks does not constitute endorsement by the Department of War of the linked websites or the information, products, or services contained therein. The Department of War does not exercise any editorial, security, or other control over the information readers may find at these locations