The Army’s Robotic Combat Vehicle effort was supposed to be the turning point for manned-unmanned formations. After a year of prototype deliveries and soldier touchpoints, what testing has actually shown is less a revolution and more a work in progress whose seams are visible from a hundred meters out. The vehicles move. The sensors work in neat slides. The autonomy falls out of mode when the terrain gets messy, and the human burden creeps back in to pick up the slack.

Industry delivered on schedule: four teams produced two platform prototypes apiece for the Army’s Phase I, and the service used those vehicles in mobility trials and soldier evaluations through 2024. That achievement matters. Hardware milestones are concrete, costly, and hard to fake. But hardware deliveries are not the same thing as operational capability. The software and networking that make an RCV tactically useful remain brittle under realistic conditions.

The clearest failure mode exposed by testing is autonomy regression. Current autonomy stacks, including the Army’s in-house work that has been referenced publicly as the Robotic Technology Kernel and later ARCS, struggle with off-road decision making. In many tests the autonomy simply drops the vehicle out of autonomous mode when confronted with unclassified obstacles like bridges, ruts, or complex terrain features. That forces direct teleoperation or human intervention and defeats the very idea of sending robots forward to reduce soldier exposure. Teams inside and outside the Army have been explicit about this shortfall.

The comms and control problem is the twin of autonomy weakness. In open terrain a control vehicle can stay hundreds to a couple thousand meters behind the robot and maintain a useful link. In forested, urban, or broken terrain the range collapses. Tests at Fort Irwin and other soldier experimentation events showed units needing more control vehicles and more personnel than originally planned to manage the same number of robots. One platoon exercise in mid 2024 restructured from two control vehicles to three to handle four RCV surrogates. That is not an academic note. It changes logistics, command loads, and risk to humans who must remain proximate to the machines they are supposed to keep out of harm’s way.

Integration is another stubborn drain on progress. The RCV program is not just about a chassis. It is a system of systems that ties sensors, weapons, autonomy, radios, mission payloads, and soldier interfaces together. Testing repeatedly shows interfaces leak: software stacks misalign, sensor feeds overload comms links, and proprietary subsystems resist rapid substitution. The Army has tried to mitigate this with an open architecture approach and by funding multiple autonomy and software pipelines. Those are sensible steps. They are also expensive and time consuming, and they expose the program to the classic trap of growing complexity faster than the team can validate it in operational conditions.

All of this points to a single blunt conclusion. The RCV prototypes are valuable and necessary experiments, but they are not yet decisive evidence that the concept can be fielded at scale without major tradeoffs. Testing has moved the needle on mobility and basic survivability concepts, but it has also highlighted how fragile the promised end state remains when confronted with real terrain, contested electromagnetic environments, and the requirement to actually reduce soldier risk rather than simply redistribute it.

If policymakers and program managers want fewer surprises in future tests they need to prioritize three things. First, fund aggressive, operationally realistic autonomy trials that punish the software with messy, ambiguous inputs rather than clean scripted courses. Second, treat communications as a primary design constraint and validate mesh, relay, and degraded links early and often. Third, stop assuming downselects or production awards will suddenly cure software immaturity. Hardware winners still need software integration in the field, and that work requires iterative soldier-in-the-loop testing under stress.

There is one final pragmatic point. The RCV program has momentum, and the vendors involved have produced impressive platforms under tight schedules. But the testing so far suggests a programmatic posture that tolerates optimistic assumptions about autonomy and comms will be punished by reality. The right outcome for soldiers is not a sexy platform chosen on paper. It is a set of capabilities that demonstrably reduce risk, integrate with existing formations without ballooning manpower needs, and degrade gracefully when networks or software fail. We are not there yet. The testing that has been done is valuable because it exposes that fragility early. The question now is whether acquisition and engineering leadership will absorb those lessons and fund the hard, unglamorous work that follows.