The Need for Task-Specific Execution in Robot Manipulation: Skill Design for Energy-Efficient Control

摘要

Many pick-and-place tasks do not require accurate position tracking throughout the whole motion, e.g. (de)palletising. However, most solutions still use a stiff industrial robot with a position controller that is overqualified for such tasks. This paper proposes to use task-specific robots and control methods to lower the hardware and energy costs. The task requirements are explicitly expressed and performed through a skill-based abstraction layer. This abstraction layer holds semantic information of the subtasks and coordinates the discrete execution of these subtasks with different control states. These states execute specific, robust continuous controllers that exploit the natural dynamics and brakes of the robot to achieve energy efficient motions. This approach was validated for a planar transportation task on a cable-driven parallel robot, and compared to a conventional approach on a serial robot. A qualitative analysis of the energy consumption indicates a lower energy consumption during certain subtasks for the proposed approach. Thus, it is shown that by relaxing the requirement of accurate trajectory tracking as well as exploiting the natural dynamics, and contact situations, significant energy savings can be obtained, compared to conventional approaches.