Ghost Has the Legs. Anduril Wants the Nervous System.

The most famous American armed ground robot of the last decade was not a fielded weapon. It was a trade-show combination.

At the Association of the United States Army exhibition in October 2021, SWORD Defense mounted its Special Purpose Unmanned Rifle—SPUR—on a Ghost Robotics Vision 60 quadruped. The photographs did what defense-exhibition photographs are designed to do: compress an emerging military argument into one unforgettable image.

A robot dog could carry a gun. The more important American story is harder to photograph. The United States is building ground robotics as a stack: one company supplies mobility, another perception or navigation, another the sensor or weapon, and another connects the machines to the wider force. Ghost Robotics and Anduril represent two different layers of that stack.

Ghost built a durable body

Philadelphia-based Ghost Robotics designed the Vision 60 as a modular quadrupedal unmanned ground vehicle, not a dedicated weapon. Its published specifications list a 51-kilogram tare weight, 10-kilogram payload, IP67 weather protection, more than three hours of continuous walking under stated conditions, and a sprint speed around 2.4 meters per second. Open interfaces support cameras, chemical sensors, manipulators, radios and other mission packages.

The autonomy is practical rather than magical. Vision 60 can avoid obstacles, assist with stairs, replay recorded routes and execute scripted missions. Ghost’s Mission Control adds lidar-based mapping and navigation in GPS-denied environments. A human can still teleoperate the platform and monitor health and video.

That mix explains the Pentagon’s early interest. Tyndall Air Force Base announced in November 2020 that security forces would begin incorporating semi-autonomous robot dogs into perimeter patrols. An Air Force follow-up stressed surveillance in difficult or unsafe terrain and cautioned that the prototypes still required development, testing and evaluation.

The Army later brought the Ghost platform into Project Convergence–Capstone 4 at Fort Irwin in March 2024. Official Army descriptions focused on reconnaissance and situational awareness alongside other robotic systems. The documented quadruped was an unarmed forward sensor in a human-machine formation—not an autonomous rifleman.

A weapon display is not a fielded weapon

SWORD’s 2021 release described SPUR as a versatile, multi-caliber, remote-operated weapon system for air, ground, maritime or fixed platforms. It selected Vision 60 because the quadruped could cross difficult terrain and adjust its posture for a stable firing position.

The distinction disappears when the photograph is reposted. Hardware integration is not a procurement record; procurement is not unit fielding; remote operation is not autonomous lethal authority.

Ghost’s commercial trajectory still shows the platform’s value. In July 2024, South Korean defense company LIG Nex1 acquired a 60 percent controlling stake for approximately $240 million, valuing Ghost at $400 million. Ghost remains Philadelphia-based but now sits inside a larger allied defense-industrial network. Its direction includes manipulation as well as sensing. A robot that can open a door, inspect an object, move hazardous material or recover equipment may be more immediately useful than one carrying a rifle.

Anduril is pursuing a different prize

Anduril’s central ground-robotics product is not a quadruped. It is an integration thesis.

In April 2024, the Army and Defense Innovation Unit selected Anduril to prototype a software architecture for the Robotic Combat Vehicle program. The objective was to let multiple vehicle types use different government and commercial autonomy packages while sharing interfaces for payloads, sensors and effectors. DIU selected Anduril and Palantir as system integrators alongside separate navigation, machine-learning and autonomy vendors.

This is the layer that turns several machines into a force. A vehicle must know where it can drive. A sensor must create a usable track. A payload must be tasked. A commander must see what the machine sees, understand its status and authorize an action. The interfaces must cooperate without trapping the government inside one vendor’s hardware.

Anduril’s Lattice platform is intended to provide that connective layer across air, land and sea. In March 2025, U.S. Special Operations Command awarded the company a three-year, $86 million mission-autonomy integration contract covering heterogeneous, multi-domain uncrewed systems. The award does not prove a fielded autonomous ground strike force. It validates demand for common software that can coordinate vehicles, sensors and effects from different manufacturers.

There is also a naming trap: Anduril’s Ghost and Ghost-X are aerial reconnaissance systems. They are unrelated to Ghost Robotics and the Vision 60 ground platform.

The RCV reversal is the lesson

The Army’s Robotic Combat Vehicle hardware effort did not proceed as planned. Fiscal 2026 budget documents state that the Army Transformation Initiative directed the service to stop RCV hardware development after approximately $225 million in base-year 2025 investment. The same language says future work would continue through the software pathway and other autonomous and unmanned initiatives.

The reversal exposes the procurement problem. A military can spend years selecting one vehicle while commercial navigation, sensors, radios and software move faster than the acquisition program. Hardware can become obsolete before a formation learns how to employ it.

Anduril’s answer is an open integration layer. Whether it works at combat scale remains unproved. In April 2026, the company reported that its Next Generation Command and Control work with the 4th Infantry Division expanded to more than 65 tactical edge nodes, over 20 applications, and dozens of data feeds, sensors and effectors during the Ivy Sting exercise series. Those are company-reported results, but they show the direction: not one signature robot, but the data mesh around every vehicle.

The market is already wider

Ghost and Anduril are only two pieces. DIU’s Ground Vehicle Autonomous Pathways effort also selected Forterra, Kodiak Robotics, Neya Systems and Overland AI for navigation; Applied Intuition and Scale AI for machine-learning infrastructure; and Palantir for integration.

By 2026, Overland’s ULTRA vehicles were operating with Army troops in demanding training events that included GPS-denied movement, logistics, reconnaissance and counter-drone work. Forterra’s autonomy stack was moving into the Marine Corps’ ROGUE Fires launcher program. Allen Control Systems was marketing weapon stations that autonomously detect and track drones while retaining an operator command to fire.

The market is converging on modularity: vehicle, autonomy, command software and effectors should be replaceable without rebuilding the entire system.

Human control is part of the architecture

The Defense Department’s 2023 autonomy directive requires autonomous and semi-autonomous weapons to permit appropriate levels of human judgment over force. It also requires realistic testing, cybersecurity, system safety and senior review for certain systems.

That does not mean every future vehicle will be remotely steered. Mobility, route selection, sensor fusion and tracking can be highly automated. It means lethal authority is a separate design question from the autonomy that gets a robot across the battlefield. SPUR was remote-operated. Current counter-drone weapons advertise operator authorization. Anduril describes human supervision over teams of autonomous systems. The phrase “armed robot” tells us little by itself.

The best software cannot rescue an unreliable vehicle. The best quadruped becomes a remote-control curiosity if it cannot function within the unit’s network and command structure. America’s ground-robot race will not be won by the most frightening trade-show image. It will be won by the system that survives mud, electronic attack, software updates, exhausted operators and the tempo of a real formation.