When Robot Safety Becomes a Perception Problem

For decades, robot safety was based on separation. Cages, light curtains, guarded workcells, and emergency stops kept machines and people apart. Physical AI changes that model. Robots increasingly share space with people, work at close range, and move through dynamic environments designed for humans.

At ImagingNext 2026, Miro Mlejnek, Sales Director at at RealSense, presents “Safety Used to Be a Mechanical Problem. In Physical AI, It’s a Perception Problem” The session examines why vision and perception are becoming foundational elements of safe robotic applications.

About Miro Mlejnek

Miro Mlejnek is Sales Director at RealSense, Inc., where he drives growth through strategic client acquisition and business development across the EMEA region. Focusing on robotics, AI, and computer vision technologies, his expertise includes distributor relations, technical sales, and marketing initiatives across computing, applied robotics, and security.

Before joining RealSense, Miro spent more than 20 years at Intel Corporation in various engineering roles. As an EMEA IoT Architect, he focused on embedded electronics and the design of mission critical systems for the aerospace, defence, and healthcare sectors.

Why perception matters for robot safety

In physical AI, safety is not defined by the robot alone. It depends on the complete application: how accurately the system senses its surroundings, how quickly it interprets changes, and how reliably it responds in real time.

This places vision and perception at the center of the safety conversation. Cameras are no longer supporting components used only for observation. They provide the spatial information robots need to navigate, interact with people, and respond to changing environments.

Building this perception layer requires several capabilities to work together. Depth accuracy affects how precisely a robot understands the position of nearby people and objects. Close-range performance and field of view determine what the system can detect around the robot. Latency influences how quickly visual information becomes an action, while environmental robustness affects performance under changing light, temperature, or operating conditions.

Edge processing and system redundancy add further layers to the architecture. Together, these capabilities form what Nielsen describes as the visual cortex of physical AI: the perception layer that transforms visual data into the spatial intelligence required for safer robotic operation.

Using the RealSense D555 or newly announced D585 as a technical framework, the session will explore how these capabilities can be applied to advanced safety applications in robotics and related fields. The focus is on the architecture and engineering requirements behind perception, rather than on the camera as an isolated product.

The future of robot safety will not be shaped only by stronger barriers. It will increasingly depend on systems that can see more clearly, interpret their surroundings more quickly, and react more intelligently.

What you’ll take away

• Why physical AI is shifting robot safety from mechanical separation toward real-time vision and perception.
• How depth accuracy, close-range performance, field of view, and latency influence safe robotic applications.
• Why environmental robustness, edge processing, and redundancy must be considered as parts of the complete perception architecture.
• How visual data becomes the spatial intelligence robots need to navigate and interact safely.
• How depth-camera architecture can support advanced safety applications in robotics and beyond.