Learning system for compliant autonomous tasks in harsh environments

CERN’s work-line is deeply related to harsh environments where radiation is its main concern towards the people working there. Therefore, a robotics team was founded to find solutions to situation where the environment is too dangerous for people to intervene. The purpose of this work is to achieve an autonomous system capable of performing repetitive tasks in structured environments that are considered harsh for humans to perform activities. Furthermore, the equipment in the surroundings are of high importance and cost, which requires the system to be complaint so that none of the equipment are damaged in case of unexpected circumstances. The integration of an online modifiable trajectory generation system known as Dynamic Movement Primitives (DMP) is used to perform specific tasks. Kinematic demonstrations along with locally weighted regression (LWR) are used to learn the trajectories and implemented in the system. A compliant behaviour is achieved through the modification of the DMPs canonical system and use of an impedance controller.