This paper presents an assisted telemanipulation framework for reaching and grasping desired objects from clutter. Specifically, the developed system allows an operator to select an object from a cluttered heap and effortlessly grasp it, with the system assisting in selecting the best grasp and guiding the operator to reach it. To this end, we propose an object pose estimation scheme, a dynamic grasp re-ranking strategy, and a reach-to-grasp hybrid force/position trajectory guidance controller. We integrate them, along with our previous SpectGRASP grasp planner, into a classical bilateral teleoperation system that allows to control the robot using a haptic device while providing force feedback to the operator. For a user-selected object, our system first identifies the object in the heap and estimates its full six degrees of freedom (DoF) pose. Then, SpectGRASP generates a set of ordered, collision-free grasps for this object. Based on the current location of the robot gripper, the proposed grasp re-ranking strategy dynamically updates the best grasp. In assisted mode, the hybrid controller generates a zero force-torque path along the reach-to-grasp trajectory while automatically controlling the orientation of the robot. We conducted real-world experiments using a haptic device and a 7-DoF cobot with a 2-finger gripper to validate individual components of our telemanipulation system and its overall functionality. Obtained results demonstrate the effectiveness of our system in assisting humans to clear cluttered scenes.

翻译：本文提出了一种辅助遥操作框架，用于从杂乱环境中抓取目标物体。具体而言，该系统允许操作者从堆叠的杂物中选择一个物体并轻松抓取，系统将协助选择最佳抓取姿态并引导操作者接近该姿态。为此，我们提出了一种物体姿态估计方案、一种动态抓取重排序策略以及一种"到达-抓取"混合力/位置轨迹引导控制器。我们将这些模块与我们先前开发的SpectGRASP抓取规划器集成，构建了一个经典的双边遥操作系统，该系统允许操作者通过触觉设备控制机器人，同时提供力反馈。针对用户选定的物体，系统首先在堆叠中识别该物体并估计其完整六自由度(DoF)位姿，随后SpectGRASP生成一组有序且无碰撞的抓取方案。基于机器人夹爪的当前位置，所提出的抓取重排序策略动态更新最佳抓取方案。在辅助模式下，混合控制器沿"到达-抓取"轨迹生成零力/力矩路径，同时自动控制机器人的取向。我们使用触觉设备和配备两指夹爪的七自由度协作机器人进行了实机实验，验证了遥操作系统的各组成部分及整体功能。实验结果证明了该系统在协助人类清理杂乱场景方面的有效性。