I’ve been looking at what changes when a Jetson Orin NX system moves from a development setup into an actual deployed vision application.
For a prototype, it’s relatively straightforward to connect a camera, run an AI model and evaluate the results.
For a production system, however, there are quite a few other things to consider:
-
How many cameras and sensors need to be connected?
-
How will multiple camera streams be synchronized?
-
What happens with long-distance camera connections?
-
How should power and thermal management be handled?
-
How do you protect the system in industrial or mobile environments?
-
How will devices be monitored and updated after deployment?
-
How much of the camera, compute and software integration needs to be customized?
This is particularly relevant for applications such as AMRs, delivery robots, warehouse automation, intelligent transportation and other edge-AI vision systems.
We’ve been working on this problem with Darsi Pro, an edge AI vision compute platform based on the NVIDIA Jetson Orin NX.
The platform combines the Jetson compute module with multi-camera connectivity, including GMSL2, sensor integration, industrial-oriented design and device-management capabilities.
What I find interesting is that the challenge isn’t really just “How much AI compute does the Jetson provide?”
It’s more about how the camera + connectivity + compute + synchronization + software + deployment pieces work together as one system.
For developers already building with Jetson Orin NX, what has been the biggest challenge when moving from a development kit to a production-ready vision system?
Is it usually camera integration, multi-camera bandwidth, synchronization, thermal/power management, mechanical design, or software deployment/maintenance?
For anyone interested in the platform, here’s the technical overview: