NVIDIA Cosmos 2: edge autonomy demonstration

This demonstration presents Shield AI Hivemind autonomy and NVIDIA's AI stack in a four-agent wide-area search mission. The speakers describe aircraft with different roles, including a communications relay and a more agile search aircraft that can fly closer to the ground. Operators direct the team's intent through text prompts rather than explicitly controlling each aircraft.

The team says it runs NVIDIA Cosmos 2 at the edge. Detections feed into a reasoning model, which assesses whether an object warrants further surveillance or whether the aircraft should resume searching elsewhere. According to the speakers, operators remain involved in every important decision and can send new prompts or commands as circumstances change.

The development discussion covers the Hivemind SDK's path from building applications to simulation, vehicle deployment and field evaluation. The team describes integrating NVIDIA and Aechelon into that loop, with NVIDIA hardware for training and large-scale simulation and Orin among the edge platforms discussed.

The operator interface receives attention too: the speakers describe presenting detection images, reasons and suggested actions while limiting information overload. This is an overview of an edge AI mission workflow, rather than a setup guide for running the software on personal hardware.