Robotic inspection is the use of a robot to position cameras and sensors, or the part itself, so that an inspection is carried out automatically and the same way every time. The robot handles the motion that a person would otherwise do by hand, such as picking a part, turning it to each face, or carrying a sensor over a large surface. Vision software, 3D sensors, scales and gauges then make the measurement and the pass/fail decision.
Robotic inspection builds on machine vision: the cameras, lighting and image analysis are the same, but a robot adds the flexibility to reach many views, handle varied parts and run without an operator. The term covers two quite different worlds, mobile robots that inspect plants and infrastructure, and fixed robot cells that inspect manufactured parts. This article explains both, then goes deeper on manufacturing, where robotic inspection is still early and growing quickly.
Two kinds of robotic inspection
When people search for robotic inspection, they usually mean one of two things. The first is asset inspection: crawlers, legged robots and drones that travel through pipelines, tanks, power plants and facilities to read gauges, measure wall thickness or find leaks, so people do not have to enter hazardous spaces. The second is part inspection: robot arms in a manufacturing cell that present parts to inspection stations, or carry sensors over parts, to check dimensions and defects at production rates.
| Asset inspection robots | Manufacturing inspection robots | |
|---|---|---|
| What moves | The robot travels to the asset | The robot moves the part or the sensor |
| Typical robots | Pipe crawlers, quadrupeds, drones, tethered vehicles | Six-axis arms, SCARA, cobots, gantries |
| Sensors | Cameras, thermal, ultrasonic, gas, lidar | 2D cameras, 3D scanners, telecentric optics, scales, gauges |
| Main goal | Keep people out of hazardous or hard-to-reach areas | Inspect 100% of parts consistently and record the data |
| Decision | Mostly data for engineers to review later | Real-time pass/fail, sort or reject at the cell |
Both share the same idea: a robot makes inspection repeatable and removes the slow, risky or tiring part of the job from people. The rest of this article focuses on manufacturing, which is where Sciotex designs and builds systems.
How a robotic inspection cell works
A manufacturing inspection cell combines a robot, a set of inspection stations and software that ties the measurements together into one decision per part. A typical sequence for small, high-tolerance parts looks like this:
Because the robot presents each face to the stations in the same pose, the measurements are far more repeatable than a person turning a part under a magnifier. Every result is logged against the part, which gives quality teams traceable data instead of a tick sheet. In the RobotVIEW robotic vision inspection system, the same cell can also count parts as it inspects them.
Robot moves the part, or robot moves the sensor
There are two basic layouts. In a part-to-sensor cell, the robot carries small parts to fixed stations. This suits parts that fit in a gripper and need several different measurements. In a sensor-to-part cell, the robot carries a camera or 3D scanner over a part that stays fixed, which suits large parts such as castings, panels or assemblies that are too big or heavy to move. Some cells combine both.
| Layout | Best for | Watch for |
|---|---|---|
| Part-to-sensor | Small parts, many measurement types, sorting into bins | Gripper design for every part variant; cycle time of each station |
| Sensor-to-part | Large or heavy parts, complex surfaces, many viewpoints | Robot path accuracy, cable management, lighting that moves with the sensor |
| Hybrid | Parts with both small features and large surfaces | More programming and calibration effort |
What robotic inspection can check
- Dimensions: length, width, diameter, height, curvature, hole position and feature location.
- Mass: weight against a tolerance band, which finds voids, missing material and mixed parts.
- Surface defects: scratches, pits, chips, cracks, dents and contamination, often classified by AI.
- 3D shape: warp, flatness and profile compared with a golden master or CAD model.
- Identity and count: part type, codes and markings, and the number of parts processed.
For more on the measurement side, see our article on automated dimensional inspection, and for 3D scanning, how a laser profilometer works.
Cycle time: a worked example
Throughput is usually the first question. Suppose a cell spends 2 seconds weighing a part, 4 seconds scanning it in 3D, 3 seconds gauging it with cameras and 3 seconds on robot moves. If every step happens one after another, each part takes 2 + 4 + 3 + 3 = 12 seconds, which is 3,600 ÷ 12 = 300 parts per hour.
If the stations work on different parts at the same time, so the scale weighs part 3 while the scanner maps part 2 and the cameras gauge part 1, the cell is limited by its slowest step instead of the sum. With a 4-second slowest station plus about a second of handling, the cell approaches one part every 4 to 5 seconds, or roughly 720 to 900 parts per hour. These figures are illustrative; real numbers depend on the part, the measurements and the accuracy needed, which is why testing sample parts first matters.
| Approach | Time per part | Parts per hour |
|---|---|---|
| Sequential stations | 12 s | 300 |
| Overlapped stations (4–5 s bottleneck) | 4–5 s | 720–900 |
Robotic inspection vs other automated inspection
A robot is not always the right answer. Fixed cameras over a conveyor are faster and simpler when every part arrives in the same orientation and only one or two views are needed. A benchtop station is cheaper for lab or low-volume work. A robot earns its place when parts need many views, several kinds of measurement, gentle handling or sorting into more than two outcomes.
| Option | Strength | Limitation |
|---|---|---|
| Robotic cell | Many views and measurements, flexible handling, sorting | Higher cost and integration effort |
| Conveyor vision | High speed, simple flow, inline with production | Limited views; parts must present consistently |
| Benchtop station | Low cost, quick to deploy, ideal for QA labs | Operator loads parts; lower throughput |
| Manual inspection | No capital cost, very flexible | Slow, subjective, fatigue-driven misses |
If you are still weighing manual checks against automation, our comparison of machine vision vs manual inspection sets out accuracy, speed and cost.
How to plan a robotic inspection project
- Define the defects and dimensions that matter, with tolerances and good and bad sample parts.
- Set the rate: parts per hour now and at peak, and how parts arrive and leave.
- Choose the layout: part-to-sensor for small parts, sensor-to-part for large ones.
- Pick the sensors for each measurement: 2D, 3D, telecentric optics, scales.
- Define outputs: pass/fail, grades, sort bins, data to MES or SPC.
- Prove it on real parts before committing to a full cell. Sciotex offers free testing of sample parts to find the right configuration.
Ready to automate a manual inspection step?
Sciotex designs and builds robotic inspection cells that measure, inspect and sort small, high-tolerance parts, built and supported in the US since 1997. See how the RobotVIEW robotic vision inspection system combines vision, 3D scanning, weighing and AI in one sequence.
Frequently asked questions
Robotic inspection is the use of a robot to position cameras and sensors, or the part itself, so an inspection is done automatically and the same way every time. It covers mobile robots that inspect plants and infrastructure, and robot cells that inspect manufactured parts.
Machine vision is the cameras, lighting and software that make the measurement. Robotic inspection adds a robot that moves the part or the sensor, so the system can reach many views, handle varied parts and sort them.
It is used for small, high-tolerance parts such as medical, electronic, aerospace and energy components, and for large parts such as castings and panels where a robot carries a scanner over the surface.
It depends on the measurements. A cell that runs four steps one after another at 12 seconds per part handles about 300 parts per hour; overlapping the stations can bring that closer to one part every 4 to 5 seconds.
Not always. Dimensions and weight are measured with conventional metrology. AI helps with variable surface defects that are hard to describe with fixed rules.
Related articles
- Machine Vision: The Complete Guide to Machine Vision Systems
- Automated Dimensional Inspection: Methods, Accuracy and How to Choose
- What Is a Laser Profilometer? How Laser Triangulation Measures Parts
- 3D Machine Vision Systems for Automated Inspection
- Machine Vision vs Manual Inspection: Accuracy, Speed, Cost
