Multi-layered grasp and manipulation system for application in uncertain environments

几十年来，机器人主要在“机器人友好”的工业环境中使用，这些环境通常是结构化的、整洁的、易于理解的，如今，机器人已经成功地应用于“人类友好”的环境，如博物馆、办公楼、百货商店和家庭。这些环境通常是不确定的、无结构的，而且大多是混乱的。抓取和操纵是机器人的核心技能。它可以用它的抓手影响它的环境。机器人抓取和操纵领域正在达到一个重要的里程碑。近年来，许多机器人已经证明，它们可以在非结构化环境中可靠地对未知物体进行基本的抓取。然而，这些机器人离人类水平的操作技能还很遥远，比如手握物体的操作，与非刚性物体的交互，以及多物体任务，比如堆叠和工具使用。因此，涉及与不确定的现实环境相互作用的高级操作对当前方法和传统方法构成了重大问题，这些方法依赖于机器人及其周围环境的精确模型。本研究计划的目标是开发一个能够在不确定环境中规划、抓取和学习如何抓取物体的集成系统。这项工作可以对从制造业到精准农业的许多应用产生重大影响。

After decades of using robots in mostly "robot-friendly” industrial environments which are typically structured, uncluttered and well understood, robots today have been successfully fielded in "human-friendly " environments such as museums, office buildings, department stores, and homes. These environments are typically uncertain, unstructured and mostly cluttered. Grasping and manipulation is a central skill for a robot. It can impact its environment using its gripper. The field of robot grasping and manipulation is reaching an important milestone. In recent years, many robots have demonstrated that they can reliably perform basic grasps on unknown objects in unstructured environments. However, these robots are still far away from being capable of human-level manipulation skills such as in-hand manipulation of objects, interactions with non-rigid objects, and multi-object tasks such as stacking and tool-usage. As such, advanced manipulations that involve interacting with uncertain real-world environments pose major problems for current approaches and traditional methods that depend on accurate models of the robot and its surroundings. The objective of this research proposal is to develop an integrated system that can plan, grasp, and learn how to grasp objects in uncertain environments. This work can have a significant impact on many applications ranging from manufacturing to precision agriculture.