We are still in the middle of moving our automations to our cloud UiPath instance. We have a particular automation that uses Citrix and ComputerVision. I made some changes in Studio which is connected to our cloud instance, and ran it unattended in our QA cloud instance. Ran fine.
But in our on-prem system some of the CV selectors are failing.
Could the fact that it was developed connected to cloud but is running against an on-prem Orchestrator be causing this?
I think the Orchestrator (cloud vs on-prem) isn’t likely the root cause here — CV selectors behave the same regardless of where Orchestrator is hosted.
The issue is more likely display differences between the two environments. Check screen resolution, DPI scaling, and Citrix color depth (32-bit vs 16-bit can change how CV sees elements).
A quick fix is re-capturing the CV anchors directly in the on-prem environment. If it still fails, I’d recommend submitting a support ticket with screenshots from both environments would help the UiPath support team investigate further.
We never had this problem before, we have been running these automations for years while developing them exactly as we are now. Only recently have we started having this issue with CV selectors failing.
Yes, it can be the cause, just not through Orchestrator itself, it is the CV server behind each environment. CV activities do not analyze the screen locally, they send the image to a Computer Vision server, and the robot usually inherits which server and key to use from the Orchestrator it is connected to, So Studio and your QA cloud robots are hitting the UiPath cloud CV service while the on prem robots are likely pointing to an on prem CV server or an older endpoint configured in your on prem Orchestrator
the part that matches your timeline is the model. The cloud CV model gets updated by UiPath continuously, an on prem CV server stays on whatever version you installed. For years both were close enough and everything matched, but once they drift apart, the descriptors you capture at design time against the newer cloud model stop matching what the older model detects at runtime. That would explain why nothing changed on your side and it still started failing, the two models just no longer read the screen the same way
to confirm you can check the CV server URL each environment actually uses, in the project settings or CV Screen Scope, and in the on prem Orchestrator CV settings, If the on prem side runs its own CV server, compare its version with cloud.. the clean fix is upgrading the on prem CV server or pointing both environments to the same endpoint if the network allows. Capturing the failing targets again on prem works as a stopgap, but the drift comes back on the next model update
Because it all happened a while ago, I can’t be sure - but I suspect our issues had something to do with our Chrome extensions being automatically upgraded to 26.10 when they shouldn’t have been.