The safety design for AI-powered vehicles and robots extended not only to the sensors inside the aircraft, but also to the cameras on the site of the facility.NVIDIA Halos connects external image recognition, safety judgment and instruction to the simulator with a closed loop.
Expand safety design from the inside to the operating environment.
NVIDIAHalos is,AIPhysical for driving vehicles and robots.AISafety design, hardware, software,AIoperating environment, and deployment lifecycle. not only failure control inside the aircraft, but alsoAIThe behavior and circumstances are included in the same safety process.
The NVIDIA Halos Outside-In Safety Blueprint for robots extends this range to fixed infrastructure cameras.The camera on the facility side and the vision AI agent capture the outside of the onboard sensors and use the information to determine the robot’s safety.
Transferring camera detection information to Safety Core judgment.
Images from external cameras are sent to AI perception and converted into detection information, such as persons and aircraft, which Safety Core receives and makes a decision to communicate to the aircraft in accordance with the set security conditions.
In the reference example, which assumes the loading of a warehouse trailer, a fixed camera monitors the positioning relationship between the worker and the autonomous forklift.Safety Core determines on the basis of its input whether the forklift can enter the trailer.
Verify the closed loop of safety judgment with SIL and HIL
The implementation is divided into the base that connects the Safety Core to an existing recognition input, the SIL that configures a closed loop with a single host, and the HIL 3 profile that includes a real-world computing device. In the SIL profile, the NVIDIA Isaac Sim moves the forklift and returns the Safety Core decision to the simulated forklift via ROS.
With the HIL profile, the Safety Core is run on the NVIDIA Thor device and can be tested for connections in configurations including real-world computing devices, however, this Blueprint is intended for early access testing, evaluation and integrated development and is not intended for security-related mainstream systems without their own certified security layer.
Connect security information from dedicated hardware to testing process.
The robot-side NVIDIA IGX Thor combines accelerated computing and functional security into one module and features a dedicated Functional Safety Island.
The NVIDIA Holoscan Sensor Bridge connects sensor data to the AI and security-related processing.
The NVIDIA Halos AI Systems Inspection Lab puts security, cybersecurity, and AI security requirements into repeatable testing.