Introduction
We have great promise of using Artificial Neural Networks (ANN) as a solution for present day robotics and for the future. Many robots have the ANN as an Artificial Intelligent Core.
The main goal of this essay is to a brief description of neural network, which can be used as an AI solution for robot evolution. Moreover, this essay notes the covered research in this point.
The essay is divided in to five sections. In section 1 I will give a short background to ANN and Robotics, which is very important for understanding the follower sections, in section 2, 3, 4 and 5 I present some way of using ANN in robotics problems. In section 5 I discuss the robot and its learning task.

1. Background
In this section, I will give definitions and basic facts about Artificial Neural Networks (ANN) and Robotic in details.
1.1 What is an Artificial Neural Network (ANN)?
According to Fausett (1994:3, ANN is an information processing system that has certain performance characteristics in common with biological neural networks. ANN consists of a number of simple processing elements called Neurons. Each neuron has an internal state call Activation.

Engelbrecht Andries states that the basic building blocks of biological neural systems are nerve cells, referred to as neurons, as ilustrated in Figure 1 , a neuron consists of a soma (cell body), dendrites and an axon.

Figure 1 The basic parts of a neuron

On the other hand, the artificial neuron consists of parts as shown in figure 1.1.2 . i.e. 1. Neuron body. 2. Weights. 3. Activation function.

Figure 1.1.2 artificial neuron
The response of one neuron is simply calculated as in figure 2.1.3 (when the binary sigmoid activation function is used)

.
Figure 1.1.3 early equations present one neuron

Using these small elements, we can build the ANN form neuron from a basic network to more complex ANNs as in figure 1.1.4.

Figure 1.1.4 Ritter, H. (1997)
The ANN has many successes in applications in many areas, like Signal Processing, pattern recognition, speech recognition, Control, and Robotics.
1.2 What are Robots?
According to Russell & Norvig (2003:901), Robots are physical agents that perform tasks by manipulating the physical world. They are equipped with effectors such as legs, wheels, joints and grippers. Robots are also equipped with sensors, which allow them to perceive their environment. Present day robotics employs a diverse set of sensors, including cameras and ultrasound or other type of sensors.

Figure 1.2.1 The Robot and the Environment.
They note some categories of robots:
Most of today''s robots fall into one of three primary categories .Manipulators, or robot arms, are physically anchored to their workplace, for example in a factory assembly line or on the International Space Station. Manipulator motion usually involves an entire chain of controllable joints, enabling such robots to place their effectors in any position within the workplace….The second category is the mobile robot. Mobile robots move about their environment using wheels, legs, or similar mechanisms….The third type is a hybrid: a mobile robot equipped with manipulators. These include the humanoid robot, whose physical design mimics the human torso.
(Russell & Norvig, 2003:901-902).
Nevertheless, in some applications, we do not need human acts with the robots, but we need mobile robots and manipulation, for example, figure 1.2.2 shows two humanoid robots and the NASA Rover on Mars.

Figure 1.2.2 the NASA Rover on Mars [3], Honda''s Asimo (left) and Sony''s QRIO (right)[4]
These categories of robots need AI control units to control the effectors, because they are the output gates for the robot, and they need map builder to locate the locomotion of the robot in the environment and do the path finding. The sensors are very important, because they are the output gates and the respond for the robot to the environment.
Figure 1.2.2 shows some of the applications we can use the ANN as a solution for Robots, for example robot control, map building, Robot vision and pattern recognition, learning tasks.

Figure 1.2.2 using neural networks as AI Core in the Robot.
2. Robot Control
Robot control is one of the most important problems in robotic science. In this section, I am going to state some examples of using ANN in Robotics as An AI control machine.
Ritter (1997) mentions that there has been much research on the implementation of both feedforward and feedback control schemes with neural networks. In addition, most of these approaches are based on multilayer perceptions or RBF networks and consider the learning of forward or inverse system models, embedded in suitable control architecture.
Research by Bianco and Nolfi (2003) suggests that we can use the Neural Controller, which is a fully connected neural network with a number of sensory neurons and motor neurons updated with the logistic function. For example, if we have a Robot with six contact sensors, nine proprioceptive sensors, and nine motors, the architecture of the neural controller can be as in figure 2.1.

Figure 2.1 the architecture of the neural controller, adapted from Bianco and Nolfi (2003)
On the other hand, a study of robot control by Salomon (1998) indicates that we can use more complex ANN models and learning algorithms, such as backpropegation or kohen network.
I note in this type of research that many researchers like to use the ANN as a Control Core but not for storing data.
3. Map Building
The main problem for Mobile robots in the environment is how to build the map in its mind, when the robot wants to move form one place to another. Maybe it seems easy to do that, but when the robot has an object in the way on the destination. It should move around the object then continue until it reaches the target.
In an article, a study by Ritter (1997) focused on neural network approaches that have been derived form the Self-Organizing map. The Self-Organizing map is a map that can extend or refresh the data. Usually, there is a variety of sensors simultaneously providing different kinds of information about the robot''s surroundings.

Figure 3.1 example of the Self-Organizing map.
The map building mechanism affects the Control Core in the robot. Therefore, it has some limitations with the hardware that limit the growing of the ANN inside the electronic machine.
On the other hand, we can use some hybrid systems between ANN and Cellular Automata Machines .

Figure 4.1 2D Cellular Automata Machines
4. Robot vision and patterns recognition

Ritter (1997) claims that Robot vision is a rich field where neural networks can make much contribution and where much work has been done in the past. In addition, he notes the importance of object recognition in the robotic applications, and its feed control unit as a sensor for information, which will help to generate better decisions.
Harrison and Koch (1998) make clear that Mobile robots are ideal candidates for using sensory modality to enhance their performance, but so far have been limited by the computational expense of processing video in real time.
Vision is not the only important sensor for the robot system, we could have a Speech or sound recognition system, and it will help to improve the speed of communication with the robot.

Figure 4.1 Objects Recognition face recognition speech or sound recognition texture recognition
Maybe in the future we will have a robot, which can smell or with artificial scene all over the body to skin similarly to humans or maybe better than the human sensors.
5. Learning Tasks

Learning tasks are not an easy target to embed within a robot. Maybe we can see many robots have learned to do some tasks, but it is similar to a recorder, we record the tasks then we ask the robot to repeat it; for example the arm robots in a car factory with the job of making and painting the cars.

Figure 5.1 tape recorder and printer

Figure 5.2 the arm robots in a car factory
However, what we want to do is create a system, which can learn tasks or human motion. Then it could repeat the task in the same situation or develop it in real time applications.
A great deal of current research work by many researchers Ross, Hart, Lawson, Webb, Prem, Poelz and Morgavi (2003). have drawn attention to the fact that mobile robotics and autonomous systems are still focused on getting a robot to learn to do some task such as pushing an object to a known location or running as fast as possible over rough ground. The learning process maybe supervised, unsupervised or a process of occasional reinforcement, but the whole aim in such work is to get the robot to achieve the task that was pre-defined by the researcher.
They point out that the robot has to have developmental architectures, and it should grow up, but growing up has some requirements in the sensors, the memory, data abstraction, Associations, Expectations and planning.
We need robots, which can learn fast, with supervised and unsupervised learning, with the ability to improve thinking through experiments.

Figure 5.3 Sunny Robot in I, Robot
Conclusion
Artificial Neural Networks (ANN) give us great promise in robotic science. In addition, the researchers focus on the point of using ANN in many ways such as toy and real problems.
Robotic learning algorithms have changed from static or classic AI to the new evolution of ANN and the genetic algorithm, and I conclude that every couple of years the definition of AI Robot changes by the new technology, AI theorem, and algorithms.
I consider if we want to have more intelligent robotic systems, we should find a way of creating very small robots as neuron cells then we can use these small robots to create Hyper-software-hardware technology in one system, or maybe if we can control some type of cells of bacteria or amoeba to build the robot and the ANN.
I hope that to develop this essay through extra research, and then present it as a paper or project with more details, applications, and results.
References

1. Albinali, F. & Davies, N. (2005). IBE2005 Papers - The IEE. Retrieved September 15, 2005 from IEE, Web site: http://www.iee.org/OnComms/PN/controlauto/FahdAlbinali-IE05.pdf

2. Bianco, R. & Nolfi, S. (2003). Evolving the Neural Controller for a Robotic Arm Able to Grasp Objects on the Basis of Tactile Sensors. AI*AI 2003:Advances in Artificial Intelligence, - 375-384.

3. Harrison, R. R. & Koch, C. (1998). A Neuromorphic Visual Motion Sensor For Real-World Robots - Harrison, Koch (ResearchIndex). -,-(-). Retrieved September 15, 2005 from http://citeseer.ist.psu.edu/harrison98neuromorphic.html

4. Nolfi, S. & Parisi, D. (1997). Neural Networks in an Artificial Life Perspective. Artificial Neural Networks-ICANN ''97, 1327 733-737.

5. Ritter, H. (1997). Self-Organizing Maps for Robot Control. Artificial Neural Networks-ICANN ''97, 1327: 675-684.

6. Ross, P., Hart, E.; Lawson, A.; Webb, A.; Prem, E.; Poelz, P. & Morgavi, G. (2003). Requirements for Gettign a Robot to Grow up. Advances in Artificial Life, (LNAI 2801), 847-856.

7. Salomon, R. (1998). Achieving Robust Behavior by Using Proprioceptive Activity Patterns - Salomon (ResearchIndex). Retrieved September 15, 2005 from http://citeseer.ist.psu.edu/salomon98achieving.html

8. Steels, L. (1995). When Are Robots Intelligent Autonomous Agents (ResearchIndex). Retrieved September 15, 2005 from MIT, Web site: http://citeseer.csail.mit.edu/17941.html

9. Tani, J. (1997). Visual Attention and Learning of a Cognitive Robot. Artificial Neural Networks-ICANN ''97, 1327 697-702.

10. Fausett, L. (1994). Fundamentals of Neural Networks. USA: Prentice-Hall.

11. Russell, S. J. & Norvig, P. (2003). Artificial Intelligence A Modern Approach. USA: Pearson Education.

12. (2003). NASA - 2003 Mars Rover. Retrieved September 21, 2005 from NASA, Web site: http://www.generation5.org/content/2004/images/asimo-qrio.jpg.

13. (-). generation5 - Anime and the Acceptance of Robotics in Japan A Symbiotic Relationship. Retrieved September 21, 2005 from -, - Web site: http://www.nasa.gov/images/content/54532main_MarsRover.jpg.