Vision-driven Tele-Operation for Robot Manipulation
It’s worth the time to acknowledge just how amazingly well we humans can perform tasks with our hands. Starting from picking up a coin to buttoning up our shirts. All these tasks for robots are still at the very forefront of robotics research & require significant interactions between vision, perception, planning & control. Becoming an expert in all of them is quite a challenge. Tele-operation augments the robot’s capability for performing complex tasks in unstructured en- vironments and unfamiliar objects with human support. It offers the robots reasoning skills, intuition, and creativity for performing these tasks in unstructured environments and unfamiliar objects. However, most Tele-operation techniques either use some sort of sensor/gloves or expensive cameras to capture the gestures of the human, making the operation bulky as well as expensive. We present a vision-based Tele-operation of the KUKA IIWA industrial robot arm that imitates in real-time the natural motion of the human operator seen from a depth camera. First, we will discuss about Wahba’s algorithm, which was used to estimate the 6-d hand pose of the operator’s hand. Wahba’s algorithm uses the predicted 3d location of the 21 hand landmarks from google’s mediapipe to estimate this 6-DoF hand pose. The hand orientation estimated above is used to tele-operate the 7-DoF KUKA IIWA manipulator in master-slave as well as in semi- autonomous mode. Then we will talk about how an object’s orientation is estimated and used in the semi- autonomous mode of operation. The object of interest for manipulation is picked by the operator’s pointing to the object in a video stream. The focused object is then detected and segmented, and the object’s pose is estimated based on its geometry of surface normals. Finally, the object’s 6-DoF pose is estimated using hand-eye calibration and robot motion is planned with a B-spline trajectory. After combining all these techniques, two modes of tele- operations for KUKA IIWA are proposed. These methods give efficient operation of robot imitating human motion as well as gesture based operation for the semi-autonomous mode of operation.