ML2R – The Competence Center Machine Learning Rhine-Ruhr

Category-agnostic segmentation of unseen objects:

Category-agnostic segmentation of unseen objects:

CNNs have the ability to enhance automation and robotic object manipulation in industry, unfortunately most of the solutions developped in research cannot be directly used in industry unless they go through a long time-consuming tailoring process. Our goal here is to eliminate tailoring and produce a generic model that is able to work and generalize well for different type of environments

Within the coarse of the project we were able to develop a CNN-based grasping pipeline that can segment and grasp unseen objects in unseen environments as shown in this video.

For that purpose we also collected a real dataset for object 6D pose detection and segmentation [https://zenodo.org/record/6103779]. For more please visit our github page [https://github.com/AnasIbrahim/image_agnostic_segmentation].

Object 6D pose detection:

Warehouses, Workshop and factories are becoming more dynamic mainly due to enhancement of mobile robots techonologies. Having a dynamic warehouse would lead to improvement in productivity but it also requires all entities to be tracked all the time. For this purpose we are developing a method for multi-view tracking of objects. The next video shows a sample of our training dataset:

A sample of our detection method working from our multi-camera system:

Gefördert durch: Bundesministerium für Bildung und Forschung
Laufzeit: bis December 2022
Ansprechpartner: Anas Gouda, M.Sc., Hazem Youssef, M.Sc.
Website: https://www.ml2r.de