Improving robot precision using jump process and granular stochastic modeling

Precision is an important issue in robotics. It can be improved by different means. The paper examines how a strategy based on external sensors and a precision stochastic modeling can be built to reduce the final position error. Therefore, the key features of the granular stochastic modeling based on both spatial resolution and joint repeatability are first outlined. Then, the paper proposes a strategy to control more precisely the robot end-effector (EE), when the process evolves in a dynamic environment. In particular, the method brings a solution to solve the difficult problem of the robot EE drift. In this context, the jump process is used to build the EE position final confidence set which is updated after each attempt. The best target is chosen in the joint space to reduce the final maximum position error in the operational space. A methodology based on an optimisation problem is detailed and allows to compute the maximum position error. The modeling is applied to a SCARA EPSON robot. Experimental results show high adequacy with the stochastic modeling: the maximum position error is reduced and finally is smaller than the repeatability index advertised by the manufacturer.