Abstract
The paper formulates an algorithm for online modification of the trajectory of a collaborative robot. A sliding mode control algorithm is developed to identify the human intent, under ignorance of the mass and inertia of the element attached to the end effector of the collaborative robot (cobot) and also under ignorance of the human behaviour while applying force to the end effector of the cobot. The sliding mode control methodology allows a human user to modify the trajectory of the cobot while in motion. The controller is developed based on the assumption that the force applied by the human agent is an outcome of a Proportional-Derivative (PD) based dynamics. Although, the human dynamics is assumed, the parameters of the assumed PD control is unknown. However, the boundedness of the human dynamics’ gain matrices lead to the need of a sliding mode controller. Subsequently, the traditional Dynamic Motion Primitives (DMPs) have been used to rapidly train the desired behaviour of a human user. Then, the local planner is modified based on the task modification. The utility of the sliding mode control algorithm and the proposed DMP approach is demonstrated by experiments on the Kinova Gen2 cobotic Arm. Also, the sliding mode control algorithm is shown to converge asymptotically under bounded human action.
