Toward a Neural-Meta Swarm for inverse kinematics, the Neural-Dragonfly Algorithm, N-DA

In robotics, inverse kinematics (IK) consist in finding the actuation solution needed for a robotic system to achieve a given work position. Traditional analytical methods may suffer from singularities and may be time consuming especially for manipulators with high degrees of freedom (DoF). This paper proses a new neural metaheuristic approach, consisting in embedding a neural network within a swarm, where the neural network will be generating a first solution which considered as the starting configuration of the swarm. The swarm will then prospect better solutions if needed while starting from the neural network initial position. The approach is called the neural-meta-swarm, N-M-swarm. This paper is a typical implementation of the approach where an artificial neural networks in embedded in the initialization process of the Dragonfly Algorithm. The ANN is trained on forward kinematics data to map end-effector positions to joint angles, while the DA optimizes the ANN's up to the needed precision. Experimental results demonstrate that this method retuned acceptable results for a 4 DoF manipulator. N-DA showed also acceptable performance for an approximated circular path planning.