As a core component of robots, the robot controller is one of the key factors affecting robot performance and influences robot development to a certain extent. Currently, due to significant advancements in artificial intelligence, computer science, sensor technology, and other related disciplines, robot research is progressing at a high level, which also places higher demands on the performance of robot controllers. For different types of robots, such as legged walking robots and articulated industrial robots, the integrated methods of control systems differ considerably, and the design schemes of controllers also vary.
A robot controller is a device that controls a robot to perform certain actions or tasks based on instructions and sensor information. It is the heart of the robot and determines the quality of the robot's performance. From the perspective of the processing method of robot control algorithms, it can be divided into two structural types: serial and parallel.
I. Serial Processing Structure
The so-called serial processing structure refers to a robot's control algorithm being processed by a serial machine. This type of controller can be further divided into the following categories based on computer architecture and control method:
1. Single CPU structure and centralized control method: This structure uses a powerful computer to implement all control functions. It was used in early robots. However, the control process requires a lot of calculations (such as coordinate transformation), so this control structure is relatively slow.
2. Two-level CPU structure and master-slave control mode: The first-level CPU acts as the master, responsible for system management, robot language compilation, and human-machine interface functions. It also utilizes its computing power to perform coordinate transformations and trajectory interpolation, periodically sending the calculation results as increments in joint motion to shared memory for the second-level CPU to read. The second-level CPU completes all digital control of joint positions. In this type of system, the two CPU buses are essentially unconnected, exchanging data only through shared memory, resulting in a loosely coupled relationship. Further distributing functions using more CPUs is very difficult.
3. Multi-CPU architecture and distributed control method
Currently, a two-tier distributed architecture with upper and lower control units is commonly used. The upper control unit is responsible for overall system management, kinematic calculations, trajectory planning, etc. The lower control unit consists of multiple CPUs, each controlling the movement of one joint. These CPUs are tightly coupled to the main control unit via a bus. This architecture significantly improves the controller's operating speed and control performance. However, these multi-CPU systems share the characteristic of employing a functionally distributed architecture tailored to specific problems, meaning each processor undertakes a fixed task. Most commercially available robot controllers worldwide currently use this architecture.
The position control section in a computer-controlled controller almost invariably employs digital position control.
All of the above types of controllers use serial machines to calculate robot control algorithms. They share a common weakness: heavy computational burden and poor real-time performance. Therefore, most of them use offline planning and feedforward compensation decoupling methods to reduce the computational burden in real-time control. When the robot is disturbed during operation, its performance will be affected, and it will be even more difficult to guarantee the accuracy required for high-speed motion.
II. Parallel Processing Structure
Parallel processing technology is an important and effective means to improve computing speed and can meet the real-time requirements of robot control. According to the literature, the research on parallel processing technology for robot controllers focuses more on parallel algorithms and their implementations for robot kinematics and dynamics.
In 1982, JYS Luh first proposed the problem of parallel processing of robot dynamics. This is because the dynamic equations of articulated robots are a set of nonlinear, strongly coupled second-order differential equations, which are very complex to calculate. Improving the calculation speed of robot dynamics algorithms also lays the foundation for realizing complex control algorithms such as the torque calculation method, nonlinear feedforward method, and adaptive control method.
One approach to developing parallel algorithms is to modify serial algorithms to make them parallel, and then map the algorithm to a parallel architecture. There are generally two methods: one is to consider a given parallel processor architecture and develop the algorithm's parallelism based on the computational model supported by the processor architecture; the other is to first develop the algorithm's parallelism, and then design a parallel processor architecture to support the algorithm, in order to achieve optimal parallel efficiency.
