Data-driven Neuro-fuzzy Multi-objective Trajectory Planning of Redundant Manipulators

In this paper, the problem of multi-objective trajectory planning is studied for redundant planar serial manipulators using a data-driven hybrid neuro-fuzzy system. This system has been developed in two major steps. First, an offline planning is performed to generate a large dataset of multi-criteria trajectories, covering mostly of the robot workspace. The later is based on a full kinematic and dynamic model of the robot, which minimizes time and energy, while avoiding singularities, and limits on joint angles, rates, accelerations, jerks and torques. An augmented Lagrangian technique is implemented on a decoupled form of the robot dynamics in order to solve the resulting non-linear constrained optimal control problem. Second, the outcomes of this first step are used to build a data-driven neuro-fuzzy inference system to learn and capture the desired dynamic behavior of the manipulator. Once this system is trained and optimized, it is used to achieve the online planning. Simulation results on a 3degrees of freedom manipulator show the effectiveness of the proposed approach.