Teleoperation Support System Robust to Uncertainties in Haptic Guidance

Abstract

When tasks need to be performed in remote, potentially inaccessible and/or hostile environments for humans, it is often more convenient to send a robot. Such environments are often only partially known, and therefore not suitable for fully autonomous robots to operate in. In these situations teleoperation can be used, which combines the problem solving capabilities of the human with the precision and durability of the robot. Robots are extremely good at performing accurate precision motions, whilst humans are best at making complex planning and decisions. Having even partial knowledge of the environment allows the provision of guidance to the operator via haptic forces. For example, this haptic guidance can assist an operator with accurate alignment of a connector for a complex connection mating task. It has been shown that haptic guidance, which is based on models of the environment, can improve task performance. However, the majority of these studies have been conducted assuming that the intended goal of the operator is known to the support system a priori and that the environment is perfectly known. In the context of space robotics, there are often multiple tasks available in the environment and the task the operator intends to execute is not known a priori. Without a goal, the support system is not able to provide the operator with haptic guidance. Moreover, due to the unstructured nature of the environment, uncertainties can result in a mismatch of the models with respect to the environment. This results in an error in the haptic guidance that is provided to the operator. The goal of this thesis is twofold: Firstly, when multiple possible goals are accessible in the environment, assistance can only be provided when the intended goal of the operator is known to the support system. Since the intention of the operator is not a physical signal that can be interpreted by the support system, it can not be communicated with the support system without performing additional actions. The operator is for example able to specify the intended goal to the support system by using a GUI or through speech. However, by utilising the control inputs of the operator to predict the intended goal, these additional actions can be skipped, which results in a more natural, seamless and faster teleoperation system. In this thesis, several prediction methods are validated and compared in simulation and by using real teleoperation data. Secondly, depending on the magnitude of uncertainty the task performance can be reduced. In order to robustify the task performance of precise manipulation tasks in the presence of uncertainty, an outer admittance loop has been implemented on an impedance controlled KUKA robot. The outer admittance control loop changes the set point of the impedance controller based on contact forces and torques. This method is implemented and validated on an experimental setup by doing peg-in-hole insertion experiments. Moreover, from this method information about the environment can be extracted in order to update the models that are used by the support system online, resulting in improved haptic guidance provided to the operator. It has been concluded that memory-based prediction methods pose as a feasible method to predict the intentions of the operator in a teleoperated reaching task. Moreover, robustness to uncertainty is not increased by making the slave more compliant. However, pseudo-admittance control poses as a feasible method to robustify task performance against uncertainty of the peg-in-hole task and can be used to reduce uncertainty by using estimated information of the environment. For future research, both methods should be validated in a human-in-the-loop study, on a teleoperation system where the operator is provided with haptic guidance, so the effects on the prediction method and the force feedback can be studied.