
Four autonomous ground vehicles searched for simulated enemy targets without relying on GPS during a US military demonstration at Fort Knox, Kentucky, on 15 October 2026.
One robot halted after detecting a possible target, leaving a human operator to authorise any attack. The Pentagon's Defense Advanced Research Projects Agency (DARPA) staged the exercise to test how robots could scout ahead of troops and report threats.
Satellite imagery guided the initial search, while onboard sensors helped the vehicles navigate and map the surrounding terrain at the Army training range.
How the Robots Found the Target
At Fort Knox's St. Vith Range, a single operator supervised four vehicles taking part in a coordinated search. After imagery indicated a simulated threat, the operator sent the team towards the suspected position.
The future is here and it’s armed. DARPA’s quadrupedal robots, once clunky prototypes hauling gear, now stalk battlefields with mounted machine guns. Twenty years after BigDog’s debut, these silent hunters clear gang held stadiums in Port au Prince. Watching one glide through rubble, soldiers at its side, isn’t just a concept video, it’s the new normal. The era of humanoids is coming. And they’re not here to haul packs anymore.
— Smokahontas (@Smokahontas2024) November 13, 2026
The lead vehicle drove along a road as the remaining three stayed behind. Halfway up a hill, it detected a potential target and stopped to await instructions.
The lead robot carried a drone inside a box on its roof, while Army photographs identified Anduril attack platforms fitted to Overland AI vehicles. DARPA said the drone could provide closer reconnaissance or, if armed, attack a target in another scenario.
Two supporting vehicles then moved forward before a simulated rocket strike hit a target about 300 yards (274 metres) away on a neighbouring hill. The fourth vehicle advanced later as the lead robot approached the hilltop.
How They Navigated Without GPS
The programme behind the vehicles is called Robotic Autonomy in Complex Environments with Resiliency, or RACER. It develops software intended to let suitably equipped ground vehicles travel through difficult terrain without GPS or pre-existing maps.
During the Fort Knox exercise, satellite imagery helped the operator choose where to send the robots, rather than steering them. Business Insider reported that the vehicles constructed maps as they moved using onboard sensors.
That distinction matters to forces operating where satellite navigation is disrupted or inaccessible. However, the public reports do not give a measure of the vehicles' accuracy, speed or reliability under deliberate electronic attack.
Jason Gregory, a DARPA technical adviser, said software allowed vehicles to choose how to move from their sensor readings. The systems could search independently and coordinate their movements across the range, but their findings still went back to a human rather than triggering an automatic strike.
Who Controlled the Simulated Strike?
Gregory said each detected target produced an image for an operator, who had to approve any engagement. The military's published account describes this as retaining human control over the decision to use force.
That approach reflects the Pentagon's directive requiring 'appropriate levels of human judgment' over the use of force. Yet the account does not disclose how reliably the robots distinguished targets, how often they raised false alarms or what information supplemented the images.
No publicly released results establish how the system would perform if communications with the operator failed. That is particularly relevant to a concept designed for contested environments where signals may be disrupted.
The exercise followed a request from US European Command to explore sending robotic scouts beyond troops' forward positions to detect threats earlier. DARPA said soldiers were expected to assess the technology in December 2026, but no combat deployment was announced.











