ADAPTIVE NEURAL NETWORK FORCE CONTROL OF ROBOT MANIPULATORS

Abstract

Many industrial applications of robot manipulators involve tasks in which the robot end-effectors are required to come into contact with environment. For successful execution of such tasks, force control is needed. In this thesis, adaptive neural networks are applied to force control of robot manipulators in an effort to improve the robustness of conventional force control schemes to parameter uncertainties and unknown model changes. Suitable neural networks are first used to parameterize the control laws and direct adaptive techniques are used to update the weights of the neural networks on-line. The introduction of neural networks removes the need to find the regressor matrix as in conventional adaptive techniques; and the use of direct adaptive methods eliminates the time-consuming off-line training of neural networks and at the same time it provides adaptive enhancements to dynamics changes. Two approaches to force control of robot manipulators are studied, namely, impedance control and constrained motion control. In the impedance control, the task space dynamics of robot manipulators are approximated using neural networks. An adaptive neural network impedance controller is designed so that the desired position trajectory can be followed during non-contact phase while the motion and force exerted on the environment are regulated according to a desired target impedance during contact phase. The proposed controller has the advantage that it does not require the inverse of Jacobian matrix which may be difficult to obtain. For the constrained motion control, an adaptive neural network controller is presented for trajectory tracking of an end-effector on a constrained surface with specified constraint force. It is shown that the motion tracking error converges to zero asymptotically whereas the force tracking error remains bounded and can be made arbitrarily small. In this thesis, another important robot control problem is also studied, namely, coordinated control of multiple manipulators. Coordination between multiple manipulators is required for tasks demanding dextrous and versatile manipulation and which are beyond the capability of a single manipulator. Two adaptive neural network control schemes are presented for coordinated control of multiple manipulators. The first scheme utilizes direct adaptive method where neural networks are used to approximate the combined dynamics of multiple manipulators. The second scheme is based on the feedback linearization approach. Adaptive neural network are added to the conventional model-based feedback linearization controller. For the neural network controllers developed in this thesis, a sliding mode control is included for closed-loop stability in the presence of neural network reconstruction errors. Rigorous stability analysis are given for the proposed adaptive neural network control schemes and it is shown that stable adaptation is assured and asymptotic tracking is achieved. Intensive computer simulations are carried out to verify the effectiveness of the proposed adaptive neural network controllers.