Beyond Top-Grasps Through Scene Completion

Abstract

Current end-to-end grasp planning methods propose grasps in the order of seconds that attain high grasp success rates on a diverse set of objects, but often by constraining the workspace to top-grasps. In this work, we present a method that allows end-to-end top-grasp planning methods to generate full six-degree-of-freedom grasps using a single RGB-D view as input. This is achieved by estimating the complete shape of the object to be grasped, then simulating different viewpoints of the object, passing the simulated viewpoints to an end-to-end grasp generation method, and finally executing the overall best grasp. The method was experimentally validated on a Franka Emika Panda by comparing 429 grasps generated by the state-of-the-art Fully Convolutional Grasp Quality CNN, both on simulated and real camera images. The results show statistically significant improvements in terms of grasp success rate when using simulated images over real camera images, especially when the real camera viewpoint is angled.

DDGC: Generative Deep Dexterous Grasping in Clutter

“…In bin-picking, objects are located in a pile and are often small and light. Methods that target bin-picking [1]- [3], [9], [10] benefit from the pile structure as objects have more stable equilibriuma and as such collisions with obstacles do not jeopardize the pile structure. In comparison, methods that target grasping in structured clutter [4], [13]-where objects are mostly larger, heavier, and packed together-need to avoid all collisions as unintended contact between the robot and objects can easily tip objects over and thus change the scene structure.…”

“…Model-based grasping in clutter assumes known object models and as such simplifies the planning problem [15]. Approaches that target model-free grasping in clutter where objects are unknown either plan grasps directly on raw sensor inputs [1]- [4], [7], [9], [13] or first estimate the shape of the unknown objects and then plan grasps [10], [14]. In this work, we estimate the shape of all the objects, as it allows us to plan grasps that are closer to the surface of the objects.…”

“…With target-driven grasp sampling methods [4] it is possible to specify the object to grasp opposed to target-free methods [1]- [3], [7], [9], [10], [13] which oftentimes simply chooses the object it can attain the best grasp on. Our method is target-driven as it segments objects and plan object-specific grasps using these segments.…”

“…1 is split into 5 sub-modules: Scene Completion, Image Encoding, Grasp Generation, Discriminator and Finger Refinement. Out of these Image Encoding and Grasp Generation are novel while the others have been used in prior work [8], [10], [30].…”

“…Scene Completion refers to the process of producing a full 3-D model of the scene by segmenting the scene and shape completing each object [10]. The method is agnostic to the segmentation algorithm.…”