• adaptive control
• optimal estimation
• pick and throw

This thesis tackles the problem of precisely executing a highly dynamic task, such as pick and place/throw of objects with different unknown inertial properties.
To solve this problem, an adaptive computed torque controller has been chosen and appropriately modified to overcome the typical problem of using the inverse of the estimated mass matrix, hence allowing intuitive control of the manipulator’s softness during its motion. Moreover, a parameter update law that guarantees zero estimation error without requiring knowledge of the dynamic values of the object was formulated. The method enables the robot to perform different operations with objects of different inertial characteristics without needing high control gains, which would make the robot stiff while maintaining a good level of precision.
To speed up the convergence and improve the accuracy in estimating the model parameters, a persistently Exciting trajectory (parameterized by sinusoids) has been obtained by solving an optimization problem. Finally, the proposed method is validated through simulations and experiments conducted on a real 7-degree-of-freedom robot in a scenario where objects of different masses are thrown in a specific location.