This work focuses on developing a complete non-linear dynamic model comprising entirely kinematic and hydrodynamic effects of Carangiform locomotion based on the Lagrange approach by adapting the parameters and behaviors of a real carp. In order to imitate biological features, swimming patterns of a real carp for forward, turning and up-down motions are analyzed by using the Kineova 8.20 software. The proportional optimum link lengths according to actual size, swimming speed, flapping frequency, proportional physical parameters and different swimming motions of the real carp are investigated with the designed robotic fish model. Three-dimensional (3D) locomotion is evaluated by tracking two trajectories in a MATLAB environment. A Reaching Law Control (RLC) approach for inner loop (Euler angles-speed control) and a guidance system for the outer loop (orientation control) are proposed to provide an effective closed-loop control performance. In order to illustrate the 3D performance of the proposed closed loop control system in a virtual reality platform, the designed robotic fish model is also implemented using the Virtual Reality Modeling Language (VRML). Simulation and experimental analysis show that the proposed model gives us significant key solutions to design a fish-like robotic prototype.

This paper presents mechatronic design and manufacturing of a biomimetic Carangiform-type autonomous robotic fish prototype (i-RoF) with two-link propulsive tail mechanism. For the design procedure, a multi-link biomimetic approach, which uses the physical characteristics of a real carp fish as its size and structure, is adapted. Appropriate body rate is determined according to swimming modes and tail oscillations of the carp. The prototype is composed of three main parts: an anterior rigid body, two-link propulsive tail mechanism, and flexible caudal fin. Prototype parts are produced with 3D-printing technology. In order to mimic fish-like robust swimming gaits, a biomimetic locomotion control structure based on Central Pattern Generator (CPG) is proposed. The designed unidirectional chained CPG network is inspired by the neural spinal cord of Lamprey, and it generates stable rhythmic oscillatory patterns. Also, a Center of Gravity (CoG) control mechanism is designed and located in the anterior rigid body to ensure three-dimensional swimming ability. With the help of this design, the characteristics of the robotic fish are performed with forward, turning, up-down and autonomous swimming motions in the experimental pool. Maximum forward speed of the robotic fish can reach 0.8516 BLs^-1 and excellent three-dimensional swimming performance is obtained.

Researchers are actively exploring advanced algorithms to enhance robots’ ability to navigate complex environments while avoiding obstacles. Four different environments were designed in the Webots simulator, including a mobile robot, a goal, a static obstacle, and one or two dynamic obstacles. The robot’s state vector was determined based on its position, the goal, and sensor variables, with all elements randomly placed in each learning and test step. A multi-layer perceptron (MLP) agent was trained for 1000 episodes in these environments using classical and fuzzy logic-based reward functions. After the training process was completed, the agents trained with the fuzzy logic-based reward function were tested for each environment. As a result of the test, while the robot’s arrival rate was 100% in the first three environments, it was measured as 91% in the fourth environment. In the last environment, the rate of crashing into a wall or dynamic obstacle was observed to be 7%. In addition, the agent trained in the fourth environment was found to successfully reach the target in multi-robot environments. The agent trained fuzzy logic-based reward function obtained the best result for four different environments. Based on these results, a fuzzy logic-based reward function was proposed to address the tuning problem of the classical reward function. It was demonstrated that a robust fuzzy logic-based reward function was successfully designed. This study contributed to the literature by presenting a reinforcement learning-based safe navigation algorithm incorporating a fuzzy logic-based reward function.

Reinforcement learning is a learning method that many creatures often unwittingly use to gain abilities such as eating and walking. Inspired by this learning method, machine learning researchers have reduced this learning method to subheadings as value learning and policy learning. In this study, the noise standard deviation of the deep deterministic policy gradient (DDPG) method, which is one of the policy learning algorithms, was examined to solve inverse kinematics of a 2 degrees-of-freedom planar robot. For this examination, 8 different functions were determined depending on the maximum value of the output of the action artificial neural network. Created artificial neural networks were trained by using these functions in 1000 iterations with 200 steps in each iteration. After the training, the statistical difference between the groups was examined and it was found that there was no statistical difference between the three best groups. For this reason, the best three groups were retrained 2500 iterations and 200 steps and tested for 100 different test scenarios after the training. After testing, the inverse kinematic equation of the 2 degrees-of-freedom planar robot with minimal errors was obtained with the help of artificial neural networks. In the light of the results, the importance of the choice of the standard deviation of noise and the correct range of selection was presented for researchers who will work in this field.

This study proposes a novel hybrid learning approach for developing a visual path-following algorithm for industrial robots. The process involves three steps: data collection from a simulation environment, network training, and testing on a real robot. The actor network is trained using supervised learning for 500 epochs. A semitrained network is then obtained at the $250^{th}$ epoch. This network is further trained for another 250 epochs using reinforcement learning methods within the simulation environment. Networks trained with supervised learning (500 epochs) and the proposed hybrid learning method (250 epochs each of supervised and reinforcement learning) are compared. The hybrid learning approach achieves a significantly lower average error (30.9 mm) compared with supervised learning (39.3 mm) on real-world images. Additionally, the hybrid approach exhibits faster processing times (31.7 s) compared with supervised learning (35.0 s). The proposed method is implemented on a KUKA Agilus KR6 R900 six-axis robot, demonstrating its effectiveness. Furthermore, the hybrid approach reduces the total power consumption of the robot’s motors compared with the supervised learning method. These results suggest that the hybrid learning approach offers a more effective and efficient solution for visual path following in industrial robots compared with traditional supervised learning.