Found UR's October announcement while looking at vision options. It's a development toolkit built around PolyScope X, not a finished bin-picking cell. I like the idea of testing recognition without assembling every part of the software stack myself. Anyone else looking at this, or is a fixed camera and a decent fixture still where you'd start?
https://www.universal-robots.com/news-and-media/news-center/universal-robots-unveils-its-ai-accelerator/
What changes between your parts? If the answer is only where they sit, I'd price the fixture first. The toolkit gets interesting when you have a perception problem you actually need to solve.
