Most surface finishing cells have 3D vision to guide the robot. So why do so many cells still fall short? Because having a sensor in the loop isn't enough if the data coming out of it is incomplete, noisy, or not accurate enough for the task. That gap between “we have vision” and “we have vision that actually works” is where most cells fail.
This quick blog explains the 4 most common causes of robotic finishing failures and how industrial-grade 3D vision systems solve them.
A burr on a machined edge can be less than a millimeter tall. A dust spot in an automotive paint coat is barely visible to the human eye. A weld bead sitting 2mm higher than nominal. These aren't edge cases; they're the details your robot needs to do its job properly.
If your 3D camera doesn't have the resolution and sub-millimeter accuracy to capture them, the robot is essentially working blind. The result? Over-grinding in some spots, missed areas in others. Mistakes at this stage are expensive ones. Precision and trueness in surface finishing aren't negotiable. It's the foundation on which everything else depends.
A lot of robot surface finishing cells use 3D vision and still struggle. Why? Having a sensor doesn't automatically mean you have good and usable data.
Parts being worked on can be very challenging. Polished steel and aluminum create highlights, while dark or matte surfaces create lowlights, leading to holes and incomplete point clouds. This leaves the robot to guess at geometries that simply weren’t captured. What you actually need is a complete point cloud. No holes, no outliers, no noise that the path planner mistakes for a real surface feature. Every gap, or false data, is a place where the robot may make mistakes.
High-mix surface finishing is a fast-growing trend in manufacturing. Where previously a robot cell would work on one part family, today it needs to handle 15 variants, each with different geometries, different tolerances, and different finishing requirements.
In a traditional cell, a new part means custom fixtures and hours to days of reprogramming. Most high-mix cells today use a robot-mounted camera. It requires the camera to be industrial-grade, compact, and lightweight. It enables the robot to scan a new part from a new angle if needed, automatically adapt its toolpath, and handle new variants without being hand-taught. Changeovers that used to take 1-2 days shrink to under an hour.
Surface finishing robots typically work in harsh environments with dust, heat and debris. Operations, like sanding, generate vibrations that may gradually degrade camera calibration. Lighting shifts between morning and afternoon. Temperatures change from one shift to the next.
Cells that aren't built for these conditions drift out of spec. The same goes for the camera. It needs to be industrial-grade, factory-calibrated across its full temperature range, and stable enough that your 2D and 3D data doesn't drift when conditions change. Robustness is what separates a cell that works on demo day from one that's still performing a year into production.
Robotic surface finishing is one of the most challenging automation problems in manufacturing. But with accurate 3D vision, adaptive toolpaths, and robust calibration, finishing cells can achieve consistent cycle times, low rework rates, and predictable production. That's why most demanding surface finishing operators, like Graymatter Robotics, use the Zivid 2+MR60.
If you want to learn more about robotic surface finishing, check out our surface finishing ebook: