Robotic picking and assembly increasingly require vision systems that can localize objects reliably, estimate depth for safe approach motions under realistic hardware constraints. This thesis presents an integrated perception-to-motion workflow developed and validated in an industrial environment at Comau, structured in three differsnt phases that move from 2D targeting to richer 3D scene understanding and finally to embedded deployment considerations. In Phase 1, a compact vision setup based on an OpenMV RT1062 RGB camera was integrated with a Time-of-Flight distance sensor (VL53L1CX/VL53L1X) on the Comau e.DO robot. Intrinsic calibration achieved a low reprojection error (RMS 0.42 px), enabling stable undistortion and repeatable centroid extraction.