The path planning problem has been studied extensively over the past decades. The goal of path planning is to find the robot motion trajectory versus time. The path planning is an important subject in robotic systems and manipulators to successfully use to their abilities for performing difficult tasks in different fields. The path planning problem is a well studied problem of robot intelligence, to which different approaches were applied, for example: neural networks(NN), potential fields, genetic algorithms(GA), particle swarm optimization(PSO). Generally speaking, these methods are classified into two categories according to the characteristics of the environment, namely the off-line global path planning and the on-line local path planning based on detecting unknown environments. In known static environments, the path can be planned offline. For unknown task spaces and environments with unpredictable changes, which require online path planning, speed and accuracy of computational algorithms are very important. Due to the redundancy in redundant robots, the path planning is regarded as an optimization problem. This optimization can be resolved dynamically or parametrically. Because of very slow and difficult solution process in dynamic optimization, this problem should be solved parametrically, in which the optimal path is generated by finding the best combination of known functions. In this thesis, the optimal paths for redundant manipulators are found by intelligent algorithms that optimize the kinematic and dynamic index through the given path of end-effector in the task space. This algorithm has been applied on a plannar 3-link robot and a cooperative robot. To validate the present work, the results of proposed algorithms have been compared with the result of the high precision search method. For different states and constraints, the optimal path is generated by soft computing techniques, including genetic algorithm, artificial bee colony optimization and particle swarm optimization. By comparing the results, it is found that the particle swarm optimization is faster and more accurate than the others. Therefore, it is used to generate data for training neural networks in online path planning. Neural networks is employed for online path planning in dynamic environments for known and unknown path of end-effector in the task space. The unknown path is calculated through the optimization problem with assuming the specified initial and target points of the end-effector. In this method, the path planning is performed in both static and dynamic environment. Two methods are proposed, where the optimal path is generated by detecting the position of obstacle in the first method and by detecting the position and measuring the velocity of obstacle in the second method and for correcting the errors, the algorithm is presented. Keywords: Path Planning, Redundant manipulator, Optimization, Soft Computing Techniques,