Abstract--Grasp detection in clutter requires the robot to reason about the 3D scene from incomplete and noisy perception. In this work, we draw insight that 3D reconstruction and grasp learning are two intimately connected tasks, both of which require a fine-grained understanding of local geometry details. We train the model on self-supervised grasp trials data in simulation. Evaluation is conducted on a clutter removal task, where the robot clears cluttered objects by grasping them one at a time. Supervision from grasp, in turn, produce better 3D reconstruction in graspable regions. Generating robust grasps from raw perception is an essential to-end on large-scale grasping datasets, either through manual task for robots to physically interact with objects in unstructured labeling [17] or self-exploration [18, 40]. This task demands the robots to reason have enabled direct grasp prediction from noisy perception. Here we consider and data-driven approaches to grasping, we investigate the the problem of 6-DoF grasp detection in clutter from 3D point synergistic relations between geometry reasoning and grasp cloud of the robot's on-board depth camera. Our goal is to learning. Our key intuition is that a learned representation predict a set of candidate grasps on a clutter of objects from capable of reconstructing the 3D scene encodes relevant geometry partial point cloud for grasping and decluttering.

In recent years, research has shown that transfer learning methods can be leveraged to construct curricula that sequence a series of simpler tasks such that performance on a final target task is improved. A major limitation of existing approaches is that such curricula are handcrafted by humans that are typically domain experts. To address this limitation, we introduce a method to generate a curriculum based on task descriptors and a novel metric of transfer potential. Our method automatically generates a curriculum as a directed acyclic graph (as opposed to a linear sequence as done in existing work). Experiments in both discrete and continuous domains show that our method produces curricula that improve the agent's learning performance when compared to the baseline condition of learning on the target task from scratch.