A Bimanual Teleoperation System of Humanoid Robots for Dexterous Manipulation

Abstract

Remote control of robots is usually used to accomplish complex tasks in unstructured environments that are inaccessible or hazardous for humans (e.g., disaster recovery). Teleoperaton of humanoid robots is commonly achieved by employing motion tracking, which simplifies the complexity of manually controlling by reducing the degrees of freedom (DOF). High-quality demonstrations are also crucial for robot learning. Despite significant research in this field, dexterous manipulation tasks remain challenging due to the complexity of control. By combining off-the-shelf techniques, we present BTO, a low-cost and intuitive bimamual teleoperation system composed of HTC Vive Trackers and data gloves to remotely control a humanoid robot. Compared with previous teleoperation systems, our approach enables operators to perform a wide range of dexterous grasping and manipulation tasks that go beyond simple pick-and-place operations. The results of experiments showcase the effectiveness and intuitiveness of our approach for remotely operating humanoid robot and collecting high-quality demonstrations, which are useful to learn visuomotor policies for fine-manipulation and dexterity tasks.