Deep Learning-based 6D pose estimation of texture less objects for Industrial Cobots

Robotics and artificial intelligence have led to a significant improvement in the automation of various industrial processes. 6D pose estimation for industrial robotic arms refers to the process of accurately determining the position and orientation of an object in 3D space relative to a robotic arm, typically in an industrial or manufacturing context. This is important for tasks such as object recognition, grasping, and manipulation, where the robot needs to know the precise position and orientation of the object in order to perform its task but the accuracy of 6D pose estimation remains a challenge. To address this, researchers created deep learning algorithms that assess the posture of objects. However, these techniques still face significant challenges in real-world industrial applications, such as lighting conditions, occlusions, and the large number of objects that need to be recognized in a short amount of time. This research aims to present an efficient method of 6D pose estimation for texture less objects. Synthetic data generation provides better generalization of the real-world data sources, making it more time and cost-effective than collecting real-world data. The major challenge is bridging the reality gap between virtual and real environments, and the data must be photorealistic for it to be used in a real time model. Domain randomization and powerful simulators are employed to replicate a real environment in order to solve these issues.