An Input-mapping Initialization Approach to Accelerate Iterative Learning of Similar Tasks

© 2019 IEEE. Iterative Learning Control (ILC) can yield superior performance for mechatronic systems that execute the same task consecutively. One major limitation of ILC however, is that the ILC algorithm has to relearn the optimal control input signal for every new task, which is time consuming. The convergence speed of the ILC can be improved by hot-starting new tasks, reusing data available from previous tasks. A hot-start is an improved initial input for a new task, which minimizes the need for additional learning. In this paper, a novel transformation-based approach to hot-start tracking or non-tracking tasks for nonlinear systems is derived. The proposed methodology is analysed in simulation examples and its effectiveness is demonstrated experimentally.status: publishe