Abstract
It is a challenging goal in robotics to make a robot grasp like a human being in a cluttered environment. Self-supervised grasp learning is one of the most promising approaches to human-like robotic grasp. However, due to inadequate feedback on grasp quality, almost the existing self-supervised grasp learning methods are coarse-grained. This paper proposes a fine-grained antipodal grasp learning (FAGL) method with augmented learning feedback. First, an indicator called antipodal degree of a grasp (ADG) is defined by a non-increasing monotonous function. ADG reflecting fine-grained grasp quality is evaluated indirectly by the destructive effect of a grasp on the environment via scene images. Next, we design a restorative sampling strategy to collect the samples of fine-grained antipodal grasps and propose a refined affordance network to generate grasp affordance maps for FAGL to decide grasp policies. Finally, in grasping actual metal workpieces, FAGL outperforms its peers in terms of grasp success rate and ADG in cluttered and adversarial scenarios by reducing the grasp effects on the surroundings. The results of extensive experiments show that our method has great potential for industrial application.
Abstract (translated)
URL
https://arxiv.org/abs/2205.05508