A Versatile Affordance Modeling Framework Using Screw Primitives to Increase Autonomy During Manipulation Contact Tasks

Recent studies utilizing Affordance Templates to perform remote contact manipulation tasks with mobile manipulators have demonstrated their usefulness for modeling complex tasks allowing robots to work in uncertain environments. These efforts largely fall into the “supervised autonomy” paradigm where a user commands high-level actions and supervises the robot’s execution, but is not responsible for the low-level execution details such as controlling the endpoint/joints, managing contact forces, or avoiding collisions. In this work, we present a novel formulation for modeling affordances that features the versatility and generality of screw theory. We also propose a generic framework and algorithm for executing manipulation contact tasks. To thoroughly evaluate the performance of the proposed formulation and framework, we performed various sets of experiments using our mobile manipulator. We showed that this framework is generic enough to successfully manipulate a variety of articulated objects. Moreover, by performing a set of manipulation tasks on a wheel valve as our case study, we demonstrated that our approach lowers task duration (about 5 times) and applied forces and torques significantly (about 2.5 and 3 times, respectively) when compared to direct teleoperation. Further, by performing 90 trials, we show robust performance of the proposed screw-based framework even when there is significant error in the estimated position and orientation (i.e., about 10 cm and 0.35 rd) of the task objects.