Shakey i Herbert
California, the beginning of the 1970s. Through rooms of the Stanford Research Institute slowly glides Shakey - the first ever mobile robot programmed with artificial intelligence algorithms that allow it to plan its actions in order to achieve its goals. Shakey the Robot is relatively tall, over 170 cm in height, and it totters when it moves, hence the name. It has a camera, ultrasonic range-finding sensors and touch sensors to enable it to know when it touches something. Shakey can navigate to the neighboring rooms, move big carton blocks scattered on the floor and move up ramps. From time to time however, e.g. when it bumps into an unexpected obstacle, it will freeze as if to ponder what to do next, and only after a while, will it take action again.
More than ten years later, and nearly 5000 km east, another bot, Herbert, makes its way across the rooms of the Massachusetts Institute of Technology (MIT) looking for empty soda cans. Once it finds one, it picks it up with its arm and takes away. It is doing great and it never freezes to think how to go about its next task - it acts, as if instinctively, without the need for deeper consideration.
The difference in the performance of the two robots is not about the superior efficiency of the electronic equipment used in the 1980s compared to that from the beginning of the 1970s; it primarily consists in a different architecture of the bot's control system.
Descartes and robotics
As for Shakey, the influence of Cartesian dualism is very clear. Descartes, a great 17th c. philosopher, claimed that senses' only function is to provide the mind with information about the environment, whereas thinking, reasoning, planning and decision making - that, for Descartes, belonged to the realm of the immaterial mind. This approach pervaded artificial intelligence in its initial period and Shakey was built on the basis of a similar principle. Information from its sensors served the purpose of creating a model of the world, and only that model was used as the basis for the process of planning, the outcome of which was a plan which was then executed. If anything unplanned occurred during the execution, for example, unknown obstacle was detected, the robot had to stop and perform the entire planning process all over again, this time based on an updated model of the world.
Oriented to behavior
It turned out after some time that that approach was ineffective and insufficient in many situations. For the world is dynamic by nature and its state may change significantly while when the robot is planning its actions, which, eventually, may lead to a situation where the plan becomes obsolete as soon as it has been created. What is more, some events like finding yourself in a dangerous situation, may require immediate, knee-jerk reaction without wasting time on contemplating different options of how to proceed.
Wherefore in the 1980s, Rodney Brooks, an MIT roboticist, decided that the key to building intelligent robots was a different approach that would mimic mechanisms used by animals who did not construct complicated models of the world. He proposed a system based on behavior. Behavior - in psychology - is nothing else than a reaction to a stimulus: whenever a stimulus triggering specific behavior occurs, the body produces a specific reaction. In architectures of this kind, actions taken by the robot are nothing else but reactions strictly linked to received stimuli, therefore, neither a model of the world nor the process of planning figure here. Of course, the robot has not one but a whole array of various behaviors at its command. Moreover, several of them may be active at once like when the robot is going towards a target and avoids an obstacle it has noticed in its way. One behavior drives approaching the target, while avoiding an obstacle is controlled by another behavior, and only when these two behaviors are coordinated, the optimal path for the robot is determined.
Toward autonomy
The latter approach contributed to a notable advance in the development of autonomous robots that was observed since the second half of the 1980s. It also brought about changes in attitude in the realm of artificial intelligence as such: from that time on, a working off line expert system that unhurriedly processes incoming information has no longer been considered an adequate metaphor of an intelligent system; the relevant metaphor now is an autonomous agent that on an ongoing basis interacts with its environment and reacts to changes thereof to achieve its goals.
Currently, hybrid architectures are the most common, which combines the two previous methods while robotics is bustling with new developments. Today, robotic vacuum cleaners clean your home on their own, robotic delivery systems provide patients with medications, and the work on autonomous cars is almost done. All these advances would not have been possible had it not been for the developments in artificial intelligence such as algorithms for image recognition, data analysis and fusion, terrain representation and mapping, localization, motion planning, and learning. Some of these algorithms date back to concepts developed for Shakey the Robot.
Students take action
These topics are also of great interest to Lodz University of Technology students who take part in the activities of the AI Section of the Student Science Club KINo affiliated with the Institute of Information Technology. They use robot simulators and Lego Mindstorm to build virtual or real robots and test a galore of algorithms for various tasks. The group is open and students who wish to give these exciting projects a try are welcome to join in.
