Every user who is familiar with the textbook of Russel / Norvig will agree, that the topic of Artificial Intelligence is a large subject which is hard to grasp. The first thing to do is to reduce AI to it’s core principle. This is not the Lisp programming language and it is not neural network. But the core idea of Artificial Intelligence is called optimal control theory. Optimal control theory was developed in the 1950s as a subpart of control theory. It is used for determine the control policy of dynamic non-linear systems.
Most robotics which are realized in the reality or should be build in the future can be realized with the help of optimal control theory. Instead of creating thinking machines which can do everything what a human can do, the more easier approach is to reduce the goal to let a robot walk or drive around. For most non-engineers a working robot is equal to Artificial Intelligence. If they see that the robot can balance similar to what a human can do, most people will think, that this is equal to Artificial Intelligence.
Sure, from an academic standpoint, AI is more than only stabilizing a biped robot. For example, all the problem of computer vision, natural language processing and symbolic reasoning are unsolved. But as a first step, it make sense to focus only on the optimal control problem.
Optimal control theory has the advantage, that many books are available about the subject. And the information in these books are providing a guideline for programming real robots. That means, if someone is an expert for control theory, this is equal that a biped robot can be realized. This makes it pretty easy to describe the research workflow.
First step is to search for around 500 existing books about optimal control theory. Second step is to read all the information carefully and third step is to build the biped robot which can walk over obstacles. All the problem which can happen in this pipeline are only minor problem. They can be solved with a better understanding of what optimal control theory is about. The only disadvantages is, that the mentioned books are hard to understand. They are using advanced mathematics and differential equations are the preconditions which are used for more advanced techniques.
In contrast to the domain of Artificial Intelligence the purpose of optimal control theory is defined very precise. It is about finding a solution for a set of differential equations. This problem is a mathematical problem. There is an objective function plus the forward model and then the mathematician has to search for a solution for this problem.
What makes optimal control theory so exciting is the fact, that this clearly defined problem has applications in many domains, especially in robotics. In comparison to well known path planning problems which are research by computer scientists, the optimal control problem is more difficult to solve. But it is not impossible to find a solution, especially with the advent of fast computers.
trollheaven 