Presenter outlines neural-network advances and transition needs as experiment ends

A presenter at the final experiment described technical upgrades intended to improve autonomous capability for soldier systems and urged planning for how to hand those capabilities to the military services as the program ends.

"Here at the last experiment, we're focusing on the scientific objectives, improving the overall architecture of the system, focusing on better perception, better global planning so we can take advantage of external information and make the overall system a better experience for the soldiers," the presenter said. They described a new neural-network architecture and changes in how operator data is processed and how models are trained to make retraining more efficient.

The presenter highlighted one new component, a "global surface net," as a tool to "do a better job of detecting negative obstacles," a capability meant to improve vehicle navigation and safety in varied terrain. They also emphasized faster iteration during field testing: "It used to be where it would take us weeks to build a new model and retrain. Since we've been out here just for 2 weeks, we've actually working on our third model right now."

As the program approaches completion, the presenter urged careful planning for transition to the services and described what the services value: "They're not really interested in buying straight up autonomy. What they're interested in, the capabilities that autonomy enables. The autonomy is really the brains. We don't care what vehicle goes on. It doesn't limit their opportunities there."

No formal decisions, procurement steps, or named sponsoring agencies were stated in the remarks provided. The briefing concluded with a focus on preparing transition paths so that the capabilities demonstrated can be adopted by the services rather than sold as standalone 'autonomy.'