Real-time grasping strategies using event camera

Abstract Robotic vision plays a key role for perceiving the environment in grasping applications. However, conventional framed-based robotic vision, suffering from motion blur and low sampling rate, may not meet automation needs of evolving industrial requirements. This paper, first time, proposes an event-based framework multiple known unknown objects cluttered scene. With advantages microsecond-level rate no event camera, model-based model-free approaches are developed objects’ respectively. The multi-view approach is used to localize scene approach, then point cloud processing utilized cluster register objects. proposed on other hand, utilizes object segmentation, visual servoing grasp planning localize, align to, targeting object. Using UR10 robot with eye-in-hand neuromorphic camera Barrett hand gripper, experimentally validated different sizes. Furthermore, it demonstrates robustness significant advantage over traditional frame-based low-light conditions.