Recorded at ROC Vox Studios in Rochester, New York, this episode of Stitchless: Circle Optics Team in Focus captures a real-time conversation with Software Manager Steve Feller and Senior Computer Vision Engineer Mitchell Baller following their return from NVIDIA GTC.
This discussion reflects what is actually shifting in how machines capture, process, and act on visual data.
You can listen to the full episode featuring Steve and Mitchell wherever you get your podcasts.
Below are three core ideas shaping the future of perception systems.
1. From Pixels to Perception
The most important shift discussed is simple, but profound.
We are moving from capturing images to understanding data.
For years, imaging systems were evaluated based on resolution, frame rate, and clarity. The goal was to produce better pictures. At GTC, that framing felt outdated.
As Steve Feller describes, images are no longer the output. They are the input.
What matters now is the system’s ability to interpret what it captures. AI is transforming visual data into something that can be analyzed, acted on, and in some cases, even reconstructed beyond what was directly observed.
This expands the definition of data. It is no longer limited to pixels. Systems are combining visual inputs with simulation, inference, and multi-modal signals to create a more complete representation of the environment.
That shift also changes how we think about truth.
If AI can enhance or reconstruct what is in a scene, then “truth” is no longer just what was measured at the sensor. It becomes contextual—dependent on how the data is processed and what the system is designed to achieve.
For Circle Optics, this reinforces a core principle.
Perception is not a feature. It is infrastructure.
If the data entering a system is incomplete or distorted, everything built on top of it inherits that limitation. The integrity of perception at acquisition becomes the foundation for everything that follows.
2. Systems Are Becoming Selective, Not Exhaustive
A second theme from the conversation is the growing need for systems to be selective.
Modern imaging platforms generate more data than humans—or even downstream systems—can reasonably process. A high-resolution panoramic system can produce the equivalent of dozens of HD displays simultaneously. No operator can meaningfully monitor that in real time.
This is where the concept of the “agentic camera” begins to take shape.
Instead of passively capturing everything, systems are starting to identify what matters. They prioritize relevant information, highlight anomalies, and enable action without requiring a human to sift through raw data.
This shift is being driven by real constraints.
Bandwidth is limited. Power is limited. Cost is always a factor.
As a result, more processing is moving to the edge—closer to where the data is captured. Systems must decide, in real time, what information is worth transmitting, storing, or acting on.
Examples discussed at GTC, including satellite imaging and robotics platforms, reinforce this reality. It is often not possible to move all captured data to a centralized location for analysis. The system itself must triage.
This is also reshaping how data is distributed.
Technologies like CloudXR and foveated streaming reflect a broader trend: not all pixels are treated equally. Higher resolution is allocated where attention is focused, while less critical areas are deprioritized.
The result is a move away from exhaustive capture and toward intentional, adaptive systems.
Not everything needs to be seen. But what matters must be seen clearly.
3. Reliability Will Define What Actually Works
The final theme is the one that ultimately determines whether any of this matters.
Trust.
As Mitchell Baller emphasizes, the most advanced system in the world is useless if it cannot be relied upon.
Agentic systems, edge processing, and AI-driven perception all introduce new capabilities—but also new risks. If a system fails to detect a critical event, or produces inconsistent outputs, trust is lost immediately.
And in many of the environments discussed—defense, robotics, autonomous systems—failure is not theoretical. It has real consequences.
This places reliability at the center of system design.
It is not enough to demonstrate capability in controlled environments or simulations. Systems must perform consistently in the real world, under constraint, with imperfect conditions.
This is where the earlier themes converge.
Perception must be accurate at acquisition. Systems must be selective in how they process and distribute data. And the entire architecture must operate in a way that can be trusted.
At Circle Optics, this shows up as a combination of optical precision, mechanical integrity, and software intelligence working together. Remove any one of those elements, and the system breaks.
Together, they enable something more than visibility.
They enable perception that can be acted on.
