Control methods for robots are divided into various types depending on the amount of control and the control algorithm. The following describes the commonly used robot control methods for each type.
I. Classification according to the control volume
According to the different space in which the control volume is located, robot control can be categorized into control in joint space and control in Cartesian space. For tandem multi-joint robots, joint space control is for the control of variables in each joint of the robot, and Cartesian space control is for the control of variables at the end of the robot. According to the different control quantities, robot control can be categorized as: position control, velocity control, acceleration control, force control, force-position hybrid control, etc. These controls can be either joint space controls or end Cartesian space controls.
The goal of position control is to make the joints or ends of the controlled robot reach the desired position. The following is an example of joint space position control for a robot. The error obtained by comparing the given value of the joint position with the current value is used as the input to the position controller, and its output is used as the given value of the joint speed control after the operation of the position controller. The joint position controller often uses the PID algorithm, can also be used fuzzy control algorithm.
First, the control acceleration of the end tool is calculated. Then, the corresponding acceleration of each joint is decomposed based on the end position, velocity and acceleration expectation, as well as the current end position, joint position and velocity, and then the control torque is computed using the kinetic equations to decompose the acceleration control, which needs to be torque controlled for each joint.
Since the joint force/torque is not easy to measure directly, and the current of the joint motor can better reflect the torque of the joint motor, the current of the joint motor is often used to indicate the current measured value of the joint force/torque. The force controller controls the joint motor to exhibit the desired force/torque characteristics based on the deviation between the desired value of the force/torque and the measured value.
It consists of two parts: position control and force control. The position control is a PI control, given as the Cartesian space position of the robot's end, and the Cartesian space position feedback of the end is obtained from the position in the joint space after kinematic computation. In the figure, T is the kinematic model of the robot and J is the Jacobi matrix of the robot. The difference between the given value of the end position and the current value is converted to a position increment in the joint space using the inverse matrix of the Jacobi matrix, which is then used as part of the joint position increment after a PI operation.
The force control is also PI controlled and is given as the Cartesian space force/torque at the end of the robot, with feedback obtained from force/torque sensor measurements. The difference between the given value and the current value of the end force/moment is converted to a force/moment in joint space using the transpose matrix of the Jacobi matrix. The force/moment in the joint space is used as another part of the joint position increment after PI operation. The outputs of the position control part and the force control part are added together as the desired value of the position increment of the robot's joints. The robot utilizes the incremental control to control the position of each of its joints. The force-position hybrid control shown in Figure 1-5 is only a simple scheme in the force-position hybrid control, which is a simplified form of the R-C (Raibert-Craig) force-position hybrid control, and some necessary corrections should be made for specific environments in practical applications.
II, Classification according to control algorithm
According to the control algorithms, the control methods of robots can be categorized into PID control, variable structure control, adaptive control, fuzzy control, neuron network control and other methods. Some literature also classifies the existing control algorithms into logic threshold control, PID control, sliding mode variable structure control, neural network control and fuzzy control. These control methods are not isolated, and are often used together in a control system.
1, PID control
In engineering practice, the most widely used regulator control law for the proportional, integral, differential control, referred to as PID control, also known as PID regulation.PID controller has been introduced nearly 70 years of history, it is simple, stable, reliable, easy to adjust and has become one of the main technologies of industrial control. When the structure and parameters of the controlled object can not be completely mastered, or do not have access to accurate mathematical models, control theory of other technologies is difficult to use, the structure and parameters of the system controller must rely on experience and field debugging to determine the application of PID control technology is most convenient.
That is, when we do not fully understand a system and the controlled object, or can not be effective means of measurement to obtain the system parameters, the most suitable for PID control technology.PID control, in practice, there are also PI and PD control.PID controller is based on the system's error, the use of proportional, integral, differential calculation of the amount of control control for control.
2, variable structure control
Variable structure control is a control scheme developed from the Soviet Union in the 1950s. The so-called variable structure control means that the control system has multiple controllers, and different controllers are used in different situations according to certain rules. The use of variable structure control has many advantages not found in other controls, and can realize the control of a class of nonlinear systems with uncertain parameters.
3, adaptive control
The so-called adaptive control, refers to the system's inputs or disturbances occurring a wide range of changes, the designed system can adaptively adjust the system parameters or control strategy, so that the output can still meet the design requirements, the basic structure as shown in Figure 2-1. Adaptive control deals with systems with "uncertainty", and seeks to reduce this uncertainty by observing the state of random variables and recognizing the system model. The result is often the achievement of certain control targets, i.e., "optimal control" is replaced by "effective control".
Adaptive control systems can be categorized into model-referenced adaptive control systems, self-correcting control systems, self-optimizing control systems, variable-structure control systems, and intelligent adaptive control systems according to their principles. Among these types of adaptive control systems, model-referenced adaptive control systems and self-correcting control systems are more mature and commonly used.
4, Fuzzy control
In fuzzy control, inputs are fuzzy quantized to become fuzzy variables, there are fuzzy variables reasoned by fuzzy rules to obtain fuzzy outputs, and after defuzzification to obtain clear outputs for control. Fuzzy control was first
proposed by Prof. Zadeh of the University of California in 1965, and in 1974, E.H. Mamdani of the United Kingdom successfully applied fuzzy control to boiler and steam engine control. Subsequently, fuzzy control has been rapidly developed in the field of control and has gained a large number of successful applications.
5, Neuron network control
Neural network control is one of the frontier disciplines in the field of automatic control developed in the late 1980s. It is a new branch of intelligent control, which opens up a new way to solve the control problems of complex nonlinear, uncertain and uncertain systems. Neural network control is a product of the combination of (artificial) neural network theory and control theory, and is a developing discipline. It brings together theories, techniques, methods and research results from disciplines including mathematics, biology, neurophysiology, brain science, genetics, artificial intelligence, computer science, automatic control, etc. Its basic structure is shown in Figure 2-2.
In the field of control, the control system with learning ability is called learning control system, which belongs to intelligent control system. Neural control is with learning ability and belongs to learning control, which is a branch of intelligent control. Neural control development so far, although only more than ten years of history, there have been a variety of control structures. Such as neural predictive control, neural inverse system control and so on.