Shared control of robot manipulators with obstacle avoidance: a deep reinforcement learning approach

The word teleoperation (which, in general, means “working at a distance”) is typically used in robotics when a human operator commands a remote agent. A teleoperated robot is often employed to substitute human beings in conditions where the latter cannot operate. A possible reason is the need to be in contact with dangerous substances, and indeed, the first robot teleoperation system was designed in the 1940s for handling nuclear and chemical materials [1] . Other reasons can be the difficulty in bringing people on missions to explore deep waters or space [2] , [3] as well as the need to work with very high precision (for example) during a surgery [4] , [5] . In certain cases, the reference provided by the human operator is not directly passed to the robot but is instead used to generate an adaptive motion. This approach is known as semiautonomous teleoperation or shared control [6] , and its aim is to reduce the workload of the human operator during the performance of a difficult task that involves controlling a robotic system.