Nvidia is extending its Halos safety platform, originally built for autonomous vehicles, into robotics. The software side is designed to enable constant monitoring of "every hardware block" and "every software library" to swiftly spot failures, according to Goel. The system also isolates safety-critical computing workloads to avoid potential interference.
Halos includes the Nvidia Holoscan Sensor Bridge, which connects sensor data with safety-related computer processing so corrupted data can be easily identified. This capability can be embedded in individual hardware components, such as a microcontroller or Field Programmable Gate Array. The package also features Nvidia's simulations for testing robots in virtual environments, along with an inspection lab program that gives robotics companies and other partners quick feedback on any safety "artifacts" that arise during development.
Adapting Halos from autonomous vehicles to robotics, however, required accommodating many different definitions of safety. Functional safety for autonomous driving generally stays the same across automotive companies and countries. A robotic vacuum cleaning a hallway, though, faces very different considerations than a robotic forklift handling heavy payloads at a warehouse loading dock. Goel said the team had to reimagine much of the foundational work to build a platform that is programmable, letting developers define custom safety functions without breaking the underlying infrastructure. "Giving flexibility can come at the cost of losing some control over the stack," he told Ars.
Custom safety functions are also essential because robots operate in complex, diverse environments. A robot moving down an aisle at six miles per hour cannot see blind spots or what is coming around the corner, so it often nearly halts before turns. As Gold put it, factories, warehouses and hospitals are unstructured environments where things can come from anywhere.
Source: Ars Technica · Summarized by HeadlinesBriefing