Runtime governance for Isaac-native fleets
NVIDIA Halos certifies how robots are built. AIRGP governs what they decide — at runtime, on every action, with cryptographic proof.
Where AIRGP sits in the Isaac stack
AIRGP enforcement runs as a ROS 2 node (thinkneo_governance) in the robot's runtime graph. It subscribes to command-layer topics — /cmd_vel, /joint_states, action goals — evaluates each intended action against the fleet policy manifest, and returns ALLOW/DENY verdicts before execution. Governance applies to decisions, not perception: the node never touches NITROS GPU pipelines, so Isaac ROS perception performance is unaffected.
Compatibility matrix
| Target | Status |
|---|---|
| Isaac ROS 3.2.x · JetPack 6 · Jetson Orin | Supported (reference target) |
| ROS 2 Jazzy | Supported |
| Isaac Sim (OmniGraph governance extension) | Beta |
| Isaac ROS 4.x · JetPack 7 · Jetson Thor | Roadmap — H2 2026 |
Validate policies in simulation, deploy with proof
NVIDIA's doctrine is simulate-before-deploy. ThinkNEO applies the same doctrine to governance: load your AIRGP policy manifest into Isaac Sim, run your simulated fleet against it, watch verdicts stream live — then deploy the identical, signed manifest to the physical fleet. Policy sim-to-real.
Benchmark methodology
Policy evaluation p99 measured at the command layer on Jetson Orin-class hardware, asynchronous to GPU inference.
Research lab or GR00T early adopter? Open protocols, no sales call — request a sandbox fleet.
Request a sandbox fleet