Tele-manipulation systems are becoming more reliant on complex local (master) devices with sophisticated control methods; hence, the cognitive load on the operator during labour intensive tasks is increasing. The operator intention detection based on task learning can lead to better robot task performance with less human effort in teleoperation for a glovebox environment (see Fig. 1). Deep Convolutional Neural Networks are proposed to learn and predict the operator intention using robotic arm and its controller spatiotemporal data. Our preliminary experimental study on glovebox tasks for nuclear applications, particularly radiation survey and object grasping, provided promising results and encouraged us for a deeper research.

UK Atomic Energy Authority, Remote Applications in Challenging Environments, Robotics and AI in Nuclear (RAIN Hub)