Portal de Periódicos da CAPES

Supervised autonomy for robotic inspection

2007; SPIE; Volume: 6561; Linguagem: Inglês

10.1117/12.718806

1996-756X

Kevin L. Moore, Johnny Mollerup Larsen, Martin Stampe Qvistgaard,

Reinforcement Learning in Robotics

Experience deploying robots for security and inspection tasks shows that often the activity of "driving" the robot interferes with the activity of observing the sensor data (often visual) collected by the robot. It has been suggested that the supervised autonomy paradigm can improve system performance. In this approach, some aspects of the robot's actions are automated, particularly motion control, freeing the operator to focus on the inspection task. In this paper we describe the laboratory-level proof-of-concept development and implementation of a semi-autonomous mode for the ODIS robot, whereby, under the direction and supervision of an operator, the robot can self-navigate in a limited structured environment while sending back video images for operator inspection. Laboratory results show the feasibility of the approach.