With the needs of social development and the expansion of robot applications, people are increasingly demanding intelligent robots.
The environment in which intelligent robots are located is often unknown and unpredictable.
In the process of researching such robots, the following key technologies are mainly involved:
1) Multi-sensor information fusion
Multi-sensor information fusion technology is a hot research topic in recent years.
It combines control theory, signal processing, artificial intelligence, probability and statistics to provide a technical solution for robots to perform tasks in complex, dynamic, uncertain and unknown environments.
There are many kinds of sensors used in robots.
They are divided into internal measurement sensors and external measurement sensors according to different purposes.
Internal measurement sensors are used to detect the internal state of the components of the robot, including:
- Specific position, angle sensor;
- Arbitrary position, angle sensor;
- Speed, angle sensor;
- Acceleration sensor;
- Tilt angle sensor;
- Azimuth sensor
External sensors include:
- Vision (measurement, sensor recognition)
- Tactile (contact, pressure, sliding sensor)
- Force sense (force, torque sensor)
- Proximity (proximity sensor, distance sensor)
- Angle sensor (tilt, direction, attitude sensor)
Multi-sensor information fusion refers to the synthesis of sensory data from multiple sensors to produce more reliable, accurate or comprehensive information.
The fused multi-sensor system can better reflect the characteristics of the detected object accurately, eliminate the uncertainty of information and improve the reliability of the information.
The fused multi-sensor information has the following characteristics: redundancy, complementarity, real-time and low cost.
At present, the multi-sensor information fusion methods mainly include:
- Bayesian estimation
- Dempster-Shafer theory
- Kalman filter
- Neural Networks
- Wavelet transform
2) Navigation and positioning
In the robot system, autonomous navigation is a core technology and a key and difficult issue in the field of robotics research.
The basic tasks of navigation have 3 points:
- Global positioning based on environmental understanding
Through the understanding of the scenes in the environment, identify human-made road signs or specific objects to complete the positioning of the robot and provide materials for the path planning;
- Target recognition and obstacle detection
Real-time detection and identification of obstacles or specific targets to improve the stability of the control system;
- Security protection
It can analyze obstacles and moving objects in the robot working environment and avoid damage to the robot.
3) Path planning
Path planning technology is an important branch of robotics research.
Optimal path planning is based on one or some optimization criteria (such as minimum work cost, shortest walking route, shortest walking time, etc.).
Find an optimal path from the start state to the target state in the robot workspace that avoids obstacles.
The path planning method can be roughly divided into two methods: traditional methods and intelligent methods.
The traditional path planning methods mainly include the following:
- Free space method
- Graph search method
- Grid decoupling method
- Artificial potential field method
Most of the global planning in robot path planning is based on the above methods, but these methods need to be further improved in path search efficiency and path optimization.
Artificial potential field method is a mature and efficient planning method in traditional algorithms.
It uses the environmental potential field model for path planning, but does not consider whether the path is optimal.
The intelligent path planning method applies artificial intelligence methods such as genetic algorithm, fuzzy logic and neural network to path planning to improve the obstacle avoidance precision of robot path planning, speed up the planning speed, and meet the needs of practical applications.
Among the most widely used algorithms are:
- Fuzzy method
- Neural Networks
- Genetic algorithm
- Q learning
- Hybrid algorithm
These methods have achieved certain research results when the obstacle environment is known or unknown.
4) Robot vision
The visual system is an important part of autonomous robots.
It is usually composed of camera, image capture card, and computer.
The work of the robot vision system includes:
- Image acquisition
- Image processing and analysis
- Output and display
The core tasks are feature extraction, image segmentation and image recognition.
How to accurately and efficiently process visual information is a key issue in the vision system.
At present, the visual information processing is gradually refined, including:
- Compression and filtering of visual information
- Environmental and obstacle detection
- Identification of specific environmental signs
- 3D information perception and processing
Among them, environmental and obstacle detection is the most important and difficult process in visual information processing.
Edge extraction is a commonly used method in visual information processing.
For general image edge extraction, such as gradient method using local data and second-order differential method, it is difficult for mobile robots that need to process images in motion to meet real-time requirements.
To this end, a method of image edge extraction based on computational intelligence is proposed, such as neural network-based methods and methods using fuzzy inference rules. In particular, Professor Bezdek J.C recently comprehensively discussed the meaning of using fuzzy logic inference for image edge extraction.
This method is specific to visual navigation, which is to integrate the road knowledge required for the robot to move outdoors, such as highway white line and road edge information, integrate into the fuzzy rule base to improve road recognition efficiency and robustness.
It has also been proposed to combine genetic algorithms with fuzzy logic.
5) Intelligent control
With the development of robotics, traditional control theory has exposed shortcomings for the inability to accurately resolve the physical objects of modeling and the ill-conditioned processes with insufficient information.
In recent years, many scholars have proposed various robotic intelligent control systems.
The intelligent control methods of the robot are:
- Fuzzy control
- Neural network control
- Fusion of intelligent control technology
The integration of intelligent control technology includes:
- Fusion of fuzzy control and variable structure control
- Fusion of neural networks and variable structure control
- Fusion of fuzzy control and neural network control
Intelligent fusion technology also includes fuzzy control methods based on genetic algorithms.
6) Man-machine interface technology
Man-machine interface technology is to study how to make it convenient for people to communicate with computers.
In order to achieve this goal, in addition to the most basic requirements of the robot controller has a friendly, flexible and convenient human-machine interface, it also requires the computer to understand the text, understand the language, speak, and even translate speak between different languages.
The implementation of these functions depends on the research of knowledge representation methods.
Therefore, the study of human-machine interface technology has both great application value and basic theoretical significance.
At present, human-machine interface technology has achieved remarkable results, and technologies such as text recognition, speech synthesis and recognition, image recognition and processing, and machine translation have begun to be put into practical use.
In addition, human-machine interface devices and interactive technologies, monitoring technologies, remote operation technologies, and communication technologies are also important components of human-machine interface technology.
Remote operation technology is an important research direction.