| The multi-objective trajectory-planning problem for robotic manipulators is considered in this thesis. In the first part, the offline-planning problem is formulated in a variation calculus framework. Optimised criteria are robot travelling time and electrical and kinetic energy, and a measure of manipulability to avoid singularities. The optimisation process is done under several constraints such as actuator limitations and passing through imposed poses. The resulting constrained non-linear and non-convex optimal control problem is solved using an augmented Lagrangian with decoupling technique. This approach has been implemented on two categories of robotic manipulators; a serial redundant manipulator and a parallel manipulator. The simulations gave very good results---as compared to only kinematic based planning and approaches based on penalty methods---in time and energy minimisation and constraints satisfaction.; The second part considers the online motion planning. Because of the limitations of conventional techniques, a data-driven neuro-fuzzy approach is developed. In a prior pre-processing step, a multi-objective trajectory planning is performed using an offline methodology to generate as many as needed trajectories covering the workspace and satisfying constraints related to robot kinematics and dynamics, task and workspace. The obtained dataset is partitioned using a subtractive clustering algorithm, initializing by the same token, for a data-driven neuro-fuzzy system parameters. Then, the neuro-fuzzy system is trained and optimized to capture the dynamic multi-objective behaviour of the robot. This system allows online motion planning with lower time consumption in a generalization phase. Simulations on a 3 DOF planar serial redundant manipulator show the robustness and high generalization capabilities of the proposed system. These results show also its advantages---for time-energy minimization, redundancy resolution, and singularity avoidance---as compared to other approaches.; Keywords. Serial Robots (SRs), Parallel Kinematic Machines (PKMs), Multi-Objective Trajectory Planning, Augmented Lagrangian, Decoupling, Kinematic Redundancy Resolution, Subtractive Clustering, Data-Driven Neuro-Fuzzy Inference Systems. |
