In this paper, an overview of the artificial neural networks is presented. Their main and popular types such as the multilayer feedforward neural network (MLFFNN), the recurrent neural network (RNN), and the radial basis function (RBF) are investigated. Furthermore, the main advantages and disadvantages of each type are included as well as the training process.

Human–robot interaction (HRI) is a broad research topic, which is defined as understanding, designing, developing, and evaluating the robotic system to be used with or by humans. This paper presents a survey on the control, safety, and perspectives for HRI systems. The first part of this paper reviews the variable admittance (VA) control for human–robot co-manipulation tasks, where the virtual damping, inertia, or both are adjusted. An overview of the published research for the VA control approaches, their methods, the accomplished collaborative co-manipulation tasks and applications, and the criteria for evaluating them are presented and compared. Then, the performance of various VA controllers is compared and investigated. In the second part, the safety of HRI systems is discussed. The various methods for detection of human–robot collisions (model-based and data-based) are investigated and compared. Furthermore, the criteria, the main aspects, and the requirements for the determination of the collision and their thresholds are discussed. The performance measure and the effectiveness of each method are analyzed and compared. The third and final part of the paper discusses the perspectives, necessity, influences, and expectations of the HRI for future robotic systems.

SUMMARY: In this paper, a multilayer feedforward neural network (NN) is designed and trained, for human–robot collisions detection, using only the intrinsic joint position sensors of a manipulator. The topology of one NN 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. The proposed approach could be applied to any industrial robot, where only the joint position signals are available. The designed NN is compared quantitatively and qualitatively with an NN, where both the intrinsic joint position and the torque sensors of the manipulator are used. The proposed method is evaluated experimentally with the KUKA LWR manipulator, which is considered as an example of the collaborative robots, using two of its joints in a planar horizontal motion. The results illustrate that the developed system is efficient and fast to detect the collisions and identify the collided link.

In this paper, an approach based on multilayer neural network is proposed for human-robot collisions detection. The neural network is trained by Levenberg-Marquardt algorithm to the dynamics of the robot with and without external contacts to detect unwanted collisions of the human operator with the robot using only the proprietary position and joint torque sensors of the manipulator. The proposed method is evaluated experimentally using the 7-DOF KUKA LWR manipulator and the results illustrate that the developed system is efficient and very fast in detecting the collisions.

In this paper, a method for variable admittance control in human-robot cooperation is proposed. A multilayer feedforward neural network is designed using the Cartesian velocity of the robot and the applied force by the operator as its inputs to modify online the virtual damping of the admittance controller. The neural network is trained online using the error backpropagation algorithm based on the error between the velocity of the minimum jerk trajectory model and the measured velocity of the robot. The performance of the proposed controller and the NN generalization ability are evaluated by conducting a point-to-point cooperative motion with multiple subjects using the KUKA LWR robot.