Industry & robotics · Navarre
Industrial manufacturer
Computer vision for industrial robotics
The client
An industrial manufacturer in Navarre that automates part of its production line with robotics. By agreement with the client, we do not publish their name.
The challenge
The robot had to interact with objects that do not always arrive in the same position or orientation. Without perception of its environment, the only alternative is to fix every part mechanically — expensive and rigid, since any change to the product means rebuilding the tooling. And shop-floor conditions are unforgiving: variable lighting, reflections and shadows mean an algorithm that works in the lab can fail on the real line. Detecting the object in the image was not enough: the system had to know where it sits in space and how it is rotated, precisely enough for the robot to act on it.
What we did
We developed a computer vision solution that identifies the object and computes both the distance to it and its rotation, so the robot can move and orient itself towards it precisely. In other words, it does not stop at «there is an object there»: it delivers the position and orientation the robot needs to approach correctly. The system was tuned with images taken under the real conditions of the line, lighting and backgrounds included, rather than in a controlled environment, because that is where it had to work.
Technologies
Python · Computer vision · OpenCV · Deep learning · Pose estimation · Robotics
Result
The robot identifies the object, knows how far away and how rotated it is, and approaches it autonomously and precisely, without mechanically positioning each part. Perception is no longer the bottleneck of the cell, and the line gains flexibility when the product changes.