What Is an Autonomous Robot?
An autonomous robot is a robot with its own necessary sensors and controllers, which can perform certain tasks independently without the input and control of external human information during operation.
- Autonomous robot is its body with various necessary sensors,
- 1. Visual system: responsible for sensing the
- Medium-sized robot
- Programming autonomous robots is a very difficult task, which is mainly manifested in the following aspects:
- Most of the behavior-based implementation methods are manually programmed for different tasks. This requires programmers to fully consider all possible situations and have a good understanding of the interaction between the robot and the environment. Some designed systems are robust to many tasks and environments, but the systems are not adaptive to the environment.
- Man cannot exhaust everything possible. Whether a robot can learn to acquire knowledge that cannot be manually coded by a programmer, such as building a map of an unknown environment, is necessary in any task whose nature changes over time.
- As robot components age, the properties of the robot's sensors and actuators may change.
- The robot executes in a multi-robot environment, and its strategy needs to be changed because it needs to respond to other robots. Learning methods can play an important role in these situations. Using a learning method to program a robot without having to tell it how to achieve its goal, just tell it what the goal is and let it meet this goal through learning. This method is undoubtedly very attractive. It is the only way to improve the adaptability of the robot and reduce the programming intensity of the programmer.
Autonomous robot classification
- Robots have different levels of behavior, and they are determined by different levels of information expression, so there are different types of learning. Brooks and Mataric concluded the following types of learning:
- Learn for calibration or adjustment of sensor parameters. This type of learning optimizes operating parameters only within a specific behavioral structure.
- Learn the real world. This type of learning builds and modifies the internal expression of the robot to the real world, which is conducive to the robot's high-level intelligent behavior such as behavior planning and decision making.
- Learn the coordination of existing behaviors. This type of learning changes the effect of existing actions on the world by coordinating the order in which existing actions are triggered and executed.
- Learn new behavior. This type of learning establishes new behavioral structures.
Practical problems of autonomous robots
- Robotics is a challenge for any learning algorithm. In the process of building an autonomous robot with learning ability and automatic knowledge acquisition, we must face many tedious questions about the real world. Cnnell and Mahadev An summarized the following issues, including robot equipment, experimental environment and learning tasks, and other aspects:
- Sensor noise . The sensors of most robots are unreliable. So the state description obtained from the sensor signal is destined to be inaccurate. Learning algorithms must be able to handle noise, so statistical smoothing techniques are often needed to overcome the effects of noise.
- The manageability of the algorithm . Robots must respond in real time to unforeseen circumstances. Therefore, the learning algorithm must not be too complicated, and each iteration of the algorithm must be completed in real time.
- Incremental algorithm. The learning algorithm must allow the robot to improve its performance while learning. Because the robot must learn while collecting experience. The data that forms the experience cannot be obtained offline.
- Limited training time . Robot training time is very limited. The learning algorithm must converge in a reasonable number of operations, because the robot needs to complete the task, and it is extremely difficult to perform millions of actions on a real robot.
- A solid source of information . All the information that a robot can obtain must ultimately come from the information extracted from its sensors or the knowledge that was forcibly encoded in the initial state. Since the state information is calculated from the sensor data, the learning algorithm must work with the limitations of the sensing device. Obviously, the ability to solve some of the issues raised above determines the success of the learning algorithm used on real robots [2 ] .
Autonomous robot learning method
- Three main learning methods are becoming more popular in the field of robotics research. They are reinforcement learning (RL), evolutionary methods (GA and EP), and methods based on artificial neural networks (ANN). Among them, the most widely used method is reinforcement learning. Reinforcement learning can be used in both cases of learning new behaviors and learning to coordinate existing behaviors. Reinforcement method is an unsupervised learning algorithm, which better fits people's psychological habits of problem solving. It is closely related to traditional artificial intelligence and optimization algorithms.