Robotic Assembly and Contact Force Control

Modern industrial robots are traditionally programmed to follow desired trajectories, with the only feedback coming from the internal position/angle sensors in the joints. The robots are in general very accurate in tracking the desired motion, and they have become indispensable in many applications, such as spot welding and painting in the automotive industry. In more complex tasks, such as physical interaction with the environment, position control of the robot might be insufficient due to the fact that it is hard, or too costly, to achieve an environment that is structured enough. This is due to inherent uncertainties, such as part variations and inexact gripping. One example of a challenging application is assembly, which is hard to accomplish using only position controlled robots. By adding a force sensor to the system, it gives the robot ability to correct for uncertainties by measuring contacts. This thesis presents a framework for force controlled robotic assembly. Assembly tasks are specified as sequences of constrained motions, where transitions are triggered by sensor events, coming either from thresholds or from more advanced classifiers. The framework is also able to explicitly deal with uncertainties, which can be estimated during execution to improve the performance. Further, a method for adaptation of force control parameters is presented, and how a singularity-free orientation representation can be used within the assembly framework. The case when no force sensor is available is also considered, and a method for estimating the external forces based on the joint control errors is presented. All methods presented are validated in experiments.