Contact modelling, parameter identification and task planning for autonomous compliant motion using elementary contacts

Still today, industrial robot tasks involving contact (compliant motion tasks) are often position-controlled and, hence, require very structured environments. This means that the robot’s tools and work pieces are accurately positioned and their dimensions are known. The aim of the presented research is to increase the robot’s autonomy by measuring its position and the contact forces during the task execution. Autonomy is indispensable for the execution of tasks with inaccurately known objects or tasks in environments that do not allow a precise positioning of the objects. Application areas include space, sub-sea or nuclear environments. The same approach is relevant for future service robots operating in environments that have been designed for humans and where even simple tasks require a lot of sensing. This work makes contributions in the areas of contact modelling, recursive parameter estimation and task plan optimisation for improved estimation accuracy. Research at our department focuses on autonomous compliant motion based on explicit geometrical contact models. The models specify the constraints, the motion degree of freedom directions and the ideal contact force directions. The models are a function of the inaccurately known geometrical parameters, i.e., the positions, orientations and dimensions of the contacting objects. This work introduces the automatic generation of complex contact models based on an elementary contact library. The geometrical parameters are estimated based on these nonlinear contact models by processing position, velocity and force measurements. This work presents an analysis of the best known Kalman Filter estimators for nonlinear systems. It additionally introduces a new Bayesian estimator able to deal with static systems (parameter estimation) with any kind of nonlinear measurement models subject to additive Gaussian measurement uncertainty. The estimator is also applicable to a limited class of dynamic systems. These contributions in online recursive estimation are also relevant to other research areas. Finally, this work presents the development of robot task plans in view of the accurate estimation of the geometrical parameters (active sensing). Active sensing is a new research area in autonomous compliant motion.