Compensation for Undefined Behaviors During Robot Task Execution by Switching Controllers Depending on Embedded Dynamics in RNN

Robotic applications require both correct task performance and compensation for undefined behaviors. Although deep learning is a promising approach to perform complex tasks, the response behaviors that are not reflected in training dataset remains challenging. In human-robot collaborative task, robot may adopt an unexpected posture due collisions other events. Therefore, robots should be able recover from disturbances completing execution of intended task. We propose method by switching between two controllers. Specifically, proposed switches learning-based model-based controllers depending on internal representation recurrent neural network learns dynamics. applied pick-and-place evaluated Experimental results simulations real demonstrate effectiveness high method.