Modular learning requires minimum code for the body of the system. In the modular system, the body of the systems code can be lightweight because every new ability connects with the system by using the physical tool.
The modular learning process is simple to explain. When the builders of the robots are connecting new servo-engines and other things to the entirety, every part of the system will tell the main system what they are, and what they should do. When the main processing unit tells the finger that it must move, the servo engine of the finger has the control algorithm for telling how it can move and how big force it can use without harming the target.
Same way in the larger entireties the new robot that is participating in the entirety tells the control system what it can use. And if the robot is taking some tool, there can be a database. That tells what kind of movements that tool needs. The database can be in the tool.
Or it can be somewhere in the cloud. The protocol, how the system uses the tool can be that. The movement is that the robot is searching the database from the tool. Then it starts to search instructions from the net. There can be an image recognition system.
That compiles the image that the robot sends to the net with the tools that are introduced there. But that system is reserved for the tools that are not containing mass memory. The safest possible solution is that if there is no mass memory the tool is only for humans. And robot asks the tool where is the mass memory. There must be a system that is protecting people against the cases that there is a program error, that the system can violate people. The databases that are stored in the tools are offering the physical learning method. Every new skill the robot has is stored in the USB stick.
In that case, every single tool is containing the instructions. And the control is that if there is something non-wanted in those databases, the system alarms the controller for searching that there is no harmful code like orders to harm people. Those movements can be shown on the screen.
If the system learns things without control, that thing can make artificial intelligence problematic. Artificial intelligence can learn things, that is not serving the purpose of the system. Artificial intelligence is a powerful tool and in closed and controlled areas operating AI doesn't require autonomous learning ability.
But if the AI goes out from the controlled areas, it requires the ability to learn things independently. The main question is, how to make that ability? The base of the AI is the database, and then the system must only store the solution for the problem in the database.
The question is how the AI is making the solution? And how it determines, what is the best solution to each problem. In computer games the self-learning artificial intelligence measures the time how long the character can stay in a certain situation. The time how the character is standing. And the effect that that character makes on to opponent determines the best of solutions.
The solution that is stored in a database depends on the mission of artificial intelligence. If the AI is controlling robots. Those solutions are the movement series that the robot requires for handling situations. If we want to make a robot, that is working in large-scale operational areas, we need a very hard and complicated AI for controlling that robot. In the cases that a robot is walking on the street and attempts to handle everyday works. The system must have certain algorithms. That is allowing it to make things. In everyday work, the robot faces many challenges, that we even cannot imagine.
When people are learning things there are many things that we are automatically learning. We know automatically what to do when we are facing things like doors. But for robots, everything like how to open the door must describe. And that is making by using the programming language. Every movement of the fingers must store and describe in the computer programs.
Autonomous learning means that when the robot is making a mistake, the control system tells what the robot should do. If the robot is walking through the red light, the control system orders it to stop. And then it tells that the green light allows going over the road. In this version, the system follows the robot by using sensors like surveillance cameras and then corrects the errors, what robots are doing.
This kind of system bases the idea that the fixed and supporting systems are cooperating with robots. So the system is forming the entirety where multiple sensors are connecting data, and then the best of the solutions is chosen. And in normal life, the best of the solutions is what takes less time and energy. When the robot has three routes in its memory, and then the robot must choose the best solution, it can simply test each of the routes. And what is the fastest is the solution what the computer chooses for the robot.
Comments
Post a Comment