Vaidio AI Vision Platform Brings Visual Reasoning to the Edge With NVIDIA Cosmos and Jetson Thor

Using NVIDIA Cosmos 3 Edge, NVIDIA Metropolis, and NVIDIA Jetson Thor, the Vaidio AI Vision platform is helping organizations bring more contextual, higher-confidence visual intelligence to edge environments where traditional server deployments are difficult.

Vision AI is becoming more useful because it is becoming more practical.

Across facilities, job sites, transportation systems, and public spaces, Vision AI already helps organizations detect important events, issue real-time alerts, and turn video into operational insight. In many cases, object detection is exactly what customers need: a faster way to identify risks, incidents, and activity across large camera networks.

In some environments, however, customers may want an additional layer of verification before an alert is escalated. A vision-based system may detect smoke, a person entering a restricted area, a vehicle stopping in an unusual place, or a worker missing required safety equipment. Visual detection is valuable on its own, but a second layer of reasoning can help reduce the volume of false alerts and determine whether the event requires immediate action.

The challenge is that this kind of AI reasoning has typically required more computing power than many edge environments can support. Remote industrial sites, mobile systems, and robot-mounted applications may not have the space, power, cooling, or network reliability needed for a server-class system.

Vaidio AI Vision Platform helps address that gap by bringing NVIDIA Cosmos 3 Edge, an reasoning vision language model (VLM) to the edge on NVIDIA Jetson Thor.

Adding Visual Reasoning to Vision AI

Most Vision AI systems begin with detection.

An object detection model can scan video and identify visible items or conditions, such as a person, vehicle, smoke, fire, safety vest, hard hat, or object of interest. But some use cases benefit from additional context. Smoke may indicate a dangerous fire, but it may also be steam, dust, or cooking activity. A person carrying an object may be carrying a tool, a package, or something that requires escalation.

This is where VLM reasoning becomes useful.

In the Vaidio workflow, the customer defines what matters. The Vaidio platform monitors the video stream for the relevant event. When the platform detects something that matches the configured condition, Vaidio uses NVIDIA Cosmos 3 Edge as an additional layer of verification, helping determine whether the event should be discarded or validated as an alert. This gives the system a way to improve confidence and support a more accurate final decision before an alert is escalated.

This architecture is also efficient because the VLM does not need to analyze every frame from every camera. Vaidio uses object detection as the first step, then sends only the relevant image or cropped region to Cosmos 3 Edge for reasoning. For example, if Vaidio detects possible smoke or fire, the system can use Cosmos 3 Edge to help determine whether the scene appears to represent a real hazard rather than steam, dust, or normal activity.

That gives customers a way to add VLM reasoning without overwhelming the edge device. It also gives operators more useful alerts: not just that something was detected, but a more informed assessment of whether the event should be escalated.

Why Edge Matters

This workflow matters where server-class infrastructure is impractical. Remote industrial sites, mobile systems, and robot-mounted applications may have limited space, power, cooling, or connectivity, but still need intelligence close to the environment being observed.

Vaidio Vision AI will use NVIDIA Cosmos 3 Edge and NVIDIA Metropolis libraries on NVIDIA Jetson Thor  brings detection, reasoning, and alert decisioning closer to where events happen. 

Early Performance Observations

Vaidio’s initial work focused on validating that NVIDIA Cosmos 3 Edge could operate inside this edge Vision AI workflow.

In early, unoptimized testing, Vaidio observed sub-second inference times even when sending full-frame 1920x1080 images to Cosmos 3 Edge, with processing times of approximately 790-890 milliseconds per image. That is already fast enough to support practical edge decisioning in many event-driven workflows.

Processing times became faster when Vaidio sent smaller areas of visual evidence. Cropped 640x480 inputs took approximately 140 milliseconds per image. Smaller 336x336 cropped object inputs took approximately 110-150 milliseconds per image.

Vaidio also observed that NVIDIA Cosmos 3 Edge running on NVIDIA Jetson Thor appears capable of supporting more than 10 cameras at 1 FPS in the evaluated workflow.

These observations reinforce the value of the combined approach: detect first, send only the relevant visual evidence to the VLM, and use reasoning selectively to support a higher-confidence decision.

What This Means for Vision AI at the Edge

For customers, the value is deployment flexibility and better decision support.

A remote industrial site can evaluate safety risks without requiring a full server deployment. A mobile inspection system can analyze conditions in the field. A robot-mounted safety application can detect potential hazards and apply local reasoning without waiting for centralized processing.

Together, NVIDIA Cosmos 3 Edge, NVIDIA Metropolis, NVIDIA Jetson Thor, and the Vaidio AI Vision Platform point toward a new class of edge Vision AI systems: practical systems that can detect what is happening, reason about why it matters when additional context is needed, and support decisions closer to where the action is.

NVIDIA Blog Backlink: https://blogs.nvidia.com/blog/siggraph-news-2026/#cosmos-3

About Vaidio

Vaidio is an AI Vision Platform that transforms video into actionable intelligence. By applying advanced AI to existing camera and video management infrastructure, Vaidio helps organizations improve safety, security, compliance, investigations, and operational effectiveness. Vaidio supports flexible deployment across on-premises, edge, cloud, hybrid, and enterprise environments, enabling customers to unlock more value from the cameras they already own.

Media Contact: marketing@vaidio.ai