Intelligent control and impedance adjustment for efficient human-robot cooperation

• Κύριος συγγραφέας: Sharkawy, Abdelnasser
• Γλώσσα: English
• Έκδοση: 2021
• Θέματα: Human-robot interaction; Collision detection method; Collided link identification; Coupled/Uncoupled neural network; Coupled system; Levenberg-Marquardt; Variable admittance controller; Multilayer feedforward neural network; Error back-propagation analysis; Minimum jerk trajectory; Jordan recurrent neural network; Real-time recurrent learning algorithm; Manipulability index; Closed kinematic chain; Genetic algorithm; Ergonomics

Human - robot Cooperation (HRC) is the research topic which aims at the complementary combination between the robot abilities and the human skills. The robots can assist humans by increasing their capabilities in precision, speed, and force. In addition, the robots could reduce the stress and the tiredness of the human operator and hence improve its working conditions. The humans could contribute to the cooperation in terms of experience, knowledge of executing a task, intuition, easy learning and adaptation, and easy understanding of control strategies. This thesis is composed from three parts: the first part about the safety of human-robot co-manipulation, the second part is about the control method for human-robot co-manipulation. In the third part, a method for improving the performance of human-robot co-manipulation is presented. Towards the safety of the human-robot co-manipulation, a multilayer feedforward Neural Network based approach is proposed for human-robot collision detection and collided link identification taking safety standards into consideration. The topology of one neural network is designed considering the coupled dynamics of the robot and trained, with and without external contacts, by Levenberg-Marquardt algorithm to detect unwanted collisions of the human operator with the manipulator and the link that is collided. Two neural networks architectures are implemented; one architecture with one hidden layer is using both the intrinsic joint position and torque sensors of the manipulator. This architecture can be applied to the collaborative robots. The other neural network architecture with two hidden layers is using only the intrinsic position sensors of the manipulator. This architecture can be applied to any robot. The experimental results prove that the developed system is considerably efficient and very fast in detecting the collisions in the safe region and identifying the collided link along the entire workspace of the joints motion of the manipulator. Separate/Uncoupled neural networks, one for each joint, are also designed and trained using the same data and their performance are compared with the coupled one. Quantitative and qualitative comparison between the both neural networks architectures is also included. In the second part of the thesis, an approach is presented for variable admittance control in human-robot co-manipulation depending on the online training of neural network. The virtual damping, the virtual inertia, or both are tuned and investigated in improving the human-robot co-manipulation. The design of the variable virtual controller is analyzed, and the choice of the neural network type and their inputs and output are justified. In case of adjusting the virtual inertia only or the virtual damping only, the multilayer feedforward neural network (MLFFNN) is selected, and the error backpropagation analysis is used to train the network. In case of adjusting both the virtual damping and the virtual inertia simultaneously, a Jordan recurrent neural network (JRNN) is designed and the real-time recurrent learning algorithm for the training process. The training of any network (whether the MLFFNN or the JRNN) is happening indirectly and based on the velocity error between the reference velocity of the minimum jerk trajectory model and the actual velocity of the robot. The proposed variable admittance controller performance is experimentally investigated, and its generalization is evaluated by conducting cooperative tasks with help of multiple subjects using the KUKA LWR manipulator under different conditions and tasks than the ones used for the neural network training. Finally, a comparative study is presented between the proposed variable controller with previous published ones. Moreover, the comparison between the variable admittance controller, where both the virtual damping and the virtual inertia are adjusted simultaneously, and variable admittance controller where only the virtual damping or the virtual inertia is adjusted. Towards improving the performance of human-robot co-manipulation, an approach for the evaluation of human-robot collaboration towards high performance is introduced and implemented. The human arm and the manipulator are modelled as a closed kinematic chain (CKC) and the proposed task performance criterion is investigated based on the manipulability index of the closed kinematic chain. The selected task is a straight motion in which the robot end-effector is guided by the human operator via an admittance controller. The best location of the selected task is determined by the maximization of the minimal manipulability along the path. Evaluation criteria for the performance are adapted considering the ergonomics literature. In the experimental set-up with a KUKA LWR manipulator, multiple subjects repeat the specified motion to evaluate the introduced approach experimentally.