Outside-In Safety blueprint boosts robot productivity and safety

Robots just got extra eyes from building cameras.

FORT Robotics and NVIDIA are rolling out an AI-driven Outside-In Safety blueprint that extends robot perception beyond onboard sensors by tapping external infrastructure cameras and AI agents. The companies say the approach, part of NVIDIA’s Halos for Robotics ecosystem, can dynamically modulate robot behavior to keep pace with real-world environments while maintaining safety guarantees. The technology was showcased this week in Chicago at the Automate conference, ahead of a joint presentation in the Humanoid Robotics Pavilion on June 23.

The core idea is pragmatic: move the sensing and decision loop outward, not just inward. Testing shows that by incorporating external sensors, a robot can operate at higher efficiency in mixed human-robot workflows without sacrificing safety. The blueprint couples the NVIDIA Outside-In Safety framework with FORT’s Trust Layer to create a safety-certified loop that uses both onboard perception and external inputs. In practice, that means a warehouse robot could adjust its speed, path planning, and task sequencing in real time as it sifts through feeds from building-mounted cameras or other fixed sensors, instead of relying solely on its own cameras and lidar.

The technical stack is explicit about compute and connectivity. The system leverages NVIDIA IGX Thor for AI compute and the Holoscan Sensor Bridge for sensor connectivity, enabling a cohesive pipeline that merges external perception with robot control. The result, according to the companies, is a more productive throughput profile: fewer conservative slowdowns in dynamic environments, fewer near-miss incidents in mixed-workflows, and a clearer path to deploying autonomous systems in facilities that already feature robust camera networks. The framing emphasizes safety as a design constraint that does not have to trade off efficiency; instead, it can be an enabler of higher operational tempo.

Industry and product teams will watch closely how this approach plays out in real-world facilities. The Outside-In Safety blueprint is positioned to convert existing infrastructure into a value lever, allowing operators to repurpose building-mounted cameras to support tasks such as inventory replenishment and truck unloading while maintaining safety standards. The combination of the FORT Trust Layer and NVIDIA’s hardware/software stack is pitched as a way to realize safety-certifiable autonomous operation in environments that are inherently more complex than a lab bench.

From a practitioner vantage point, two to four concrete takeaways stand out. First, there is a clear incentive to leverage existing infrastructure to boost ROI, but that hinges on reliable, well-integrated camera networks and robust calibration between external sensors and onboard perception. Second, resilience matters: external sensors introduce new failure modes, such as occlusion, calibration drift, or connectivity hiccups, that must be accounted for in safety cases and fail-safe logic. Third, certification becomes practical only if the external sensing chain is treated as an integral part of the safety architecture, not an optional add-on; the blueprint’s emphasis on safety-certifiable operation signals this trend. Fourth, facility-scale deployment will require standardization across camera types, placements, and data formats to avoid bespoke integrations with every site.

As the Automate demo unfolds and the June 23 presentation approaches, the industry will scrutinize whether Outside-In Safety translates into reliable gains in throughput at scale or remains a compelling proof of concept. If the blueprint can consistently fuse external perception with onboard control without compromising latency or reliability, it marks a substantive shift in how factory floors and warehouses approach autonomous work.