Motion Planning by Integration of Multiple Policies for Complex Assembly Tasks

Robotic assembly has been an active area of manipulation research for several decades. However, almost all assembly tasks, especially complex ones, still need to be performed manually in industrial manufacturing. The difficulty in planning appropriate motion is a major hurdle to robotic assembly. In assembly tasks, manipulated objects come into contact with the environment. Thus, force control techniques are required for successfully achieving operations by regulating the reaction forces and dealing with uncertainties such as the position errors of robots or objects. Under force control, a robot’s responsiveness to the reaction forces is determined by force control parameters. Therefore, planning assembly motions requires designing appropriate force control parameters. Many studies have investigated simple assembly tasks such as peg-in-hole, and some knowledge of appropriate force control parameters for the tasks has been obtained by detailed geometric analysis (Whitney, 1982). However, the types of parameters that would be effective for other assembly tasks are still unknown. Here, it should be noted that the efficiency is always required in industrial application. Therefore, force control parameters that can achieve successful operations with a short time are highly desirable. However, it is difficult to estimate the cycle time, which is the time taken to complete an operation, analytically. Currently, designers have to tune the control parameters by trial and error according to their experiences and understanding of the target tasks. In addition, for complex assembly, such as insertion of complex-shaped objects, a robot's responsiveness to the reaction forces is needs to be changed according to the task state. Since tuning force control parameters with determining task conditions for switching parameters by trial and error imposes a very heavy burden on designers, complex assembly has been left for human workers. Several approaches to designing appropriate force control parameters have been presented. They can be classified as follows: (a) analytical approaches, (b) experimental approaches, and (c) learning approaches based on human skill. In the analytical approaches, the necessary and sufficient conditions for force control parameters that will enable successful operations are derived by geometric analysis of the target tasks (e.g., Schimmels, 1997, Huang & Schimmels, 2003). However, the analytical approaches cannot be utilized for obtaining the parameters to achieve operations efficiently since the cycle time cannot be 5