This article discusses the impact of the optical properties of materials and illuminance on the performance of an AI-based vision system in the bin-picking process. The process is carried out by an ABB GoFa CRB 15000-10/1.52 collaborative robot equipped with a Cambrian vision system. As part of the research, a dedicated test stand was developed. The study was conducted using parts produced via 3D printing from PLA and PETG materials (representing matte and reflective surface properties), analyzing the impact of illuminance, surface properties, partial occlusion of objects, and changes in their geometry on detection performance. The results showed that increasing the illuminance from 100 lx to 350–500 lx reduced the prediction generation time from approximately 0.9 s to approximately 0.3 s. It was noted that higher detection efficiency was achieved for workpieces with matte surfaces compared to reflective surfaces. The developed system correctly detected potential collisions and recognized the objects it had been trained to identify, even when they were partially obscured. Research has shown that a change in the part's geometry prevented its proper identification.