- https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9982158
-
Abstract— Previous studies on robot teleoperation have focused on controlling the robot’s actions, from controlling its joints to specifying the pose of its end-effector. However, these interfaces are often designed for skilled operators with extensive knowledge of robotics. In order to make teleoperation more accessible to non-experts, we propose a new framework called “Scene Editing as Teleoperation” (SEaT). The key idea behind SEaT is to transform the traditional robot-centric interface into a scene-centric interface, where users manipulate digital representations of real-world objects to specify the task’s goal instead of directly controlling the robot. This allows users to perform teleoperation without needing expert knowledge of the robot’s hardware. To achieve this, we use a category-agnostic scene-completion algorithm to create a virtual representation of the real-world workspace, which includes unknown objects that can be manipulated. We also use an action-snapping algorithm to refine the user’s input before generating the robot’s action plan. To train these algorithms, we created a large-scale dataset of object-kit pairs that simulate real-world object-kitting tasks. Our experiments, both in simulation and with a real-world system, show that our framework improves the efficiency and success rate of 6DoF kit-assembly tasks. A user study also demonstrates that participants using the SEaT framework achieve a higher task success rate and report a lower subjective workload compared to an alternative robot-centric interface.
focus
-
While the idea of task-driven teleoperation has been explored in simple scenarios such as point-goal navigation [18] or manipulation with known objects [19], there is still a research gap in enabling precise and efficient task specification for complex assembly tasks with unknown object parts. This paper aims to address this research question.
-
It seems that they are doing various things, such as capturing real point clouds and refining user input.
-
The experimental results show that SEaT reduces mental load and improves operation speed.