Unlocking the Secrets of Occluded Objects: AI Achieves New Level of Perception in Robot Vision

Robot vision can now identify objects that are occluded by using a new method

Artificial intelligence systems must make estimates based on only the visible portions of objects when they encounter scenes with objects that are not fully visible. The partial information can lead to errors in detection, and large amounts of training data are required to recognize these scenes correctly. Researchers at the Gwangju Institute of Science and Technology developed a framework to allow robot vision detect objects in the same manner as we do.

Robotic vision is a sophisticated technology that has been used in demanding and complex tasks such as autonomous driving, object manipulation, and more. It still struggles to recognize individual objects when there are cluttered scenes with objects partially or fully hidden behind other objects. When dealing with scenes like this, robotic vision systems will be trained to identify the occluded objects based on only their visible parts. This training can be tedious and requires a large number of objects.

This problem was faced by Associate Professor Kyoobin Lee and Seunghyeok back, a Ph.D. candidate from the Gwangju Institute of Science and Technology in Korea when they developed an artificial intelligence system that would identify and sort items in cluttered scenes. We expect robots to be able to identify and manipulate objects that they have never seen before or were not trained to recognize. \”In reality, we must manually collect and label each data point, as the generalizability and quality of deep neural networks is highly dependent on the quantity and quality of the training dataset,\” explains Mr. Back.


Leave a Reply

Your email address will not be published. Required fields are marked *