3203. Introduction to Artificial Intelligence

Robot and Agent


1. Action

While "perception" can be taken as an extension of input, "action" can be taken as an extension of output. An intelligent system not only can send messages to other systems, but also can directly cause changes in its external and internal environment.

An intelligent system normally has a set of basic actions implemented by its "effectors" or "actuators", so the problems related to action are when to take which action, as well as how to construct plans or programs recursively, so as to improve the efficiency of the system.

Several related AI techniques have been introduced previously:

There are also several related mathematical fields, such as decision theory, game theory, and operation research.


2. Robot

Intuitively speaking, what distinguishes a robot from an ordinary computer system is its ability to directly interact to its environment, using sensors and actuators.

There are many robots that are fully programmed or remotely controlled. What makes intelligent robots special are their flexibility and autonomy. An intelligent robot does not have to look or to behave like a human being, though there are good reasons to built humanoid robots.

Robotics is a field where many AI technologies can be applied. An intelligent robot may have the capability of perceiving, learning, reasoning, language using, and so on. Even so, the current focus of robotics is action, as well as the other issues that are directly related to it.

Robotics has a higher demand on realistic and practical models than other AI fields, since a robot needs to directly interact to its environment, so it usually has to make response in real time in an uncertain and dynamic environment.

Different approaches have been explored in building robots.

Planning-based architeture are top-down, and require the robot to perform a sequence of processing sense-plan-act steps (e.g., "combine the sensory data into a map of the world, then use the planner to find a path in the map, and finally send steps of the plan to the robot's wheels"). This process may be repeated at different levels of detail. Example: Shakey the robot and the assignment on planning ("The goal is to build a plan for the robot to open the door with the key."). Issues: response time, flexibility, and robustness.

Reactive architeture achieves rapid real-time responses by embedding the robot's controller in a collection of preprogrammed, concurrent condition-action rules (e.g., "if bumped, stop", "if stopped, back up"). Subsumption architecture provides a layered approach to assembling reactive rules into complete control systems from the bottom up. Rules, and layers of rules, are added incrementally; lower layers can function independently of the higher ones, and higher ones utilize the outputs of the lower ones, but do not override them. Example: Roomba and Cog. Issues: changing goals and learning ability.

Hybrid architecture attempts a compromise between bottom-up and top-down by employing a reactive system for low-level control and a planner for high-level decision making. Also called "three-layer architectures", they separate the control system into three communicating but independent parts: (a) the planner, (b) the reactive system, and (c) the intermediate module, which reconciles the different time-scales and representations used by the other two and any conflicts between their outputs. Example: AuRA. Issues: flexibility, complexity, and efficiency.

Representative projects: AAAI Topics on Robots, the Robot Hall of Fame


3. Agent

The term "agent" has been used in AI with several different senses, though roughly speaking an agent can be seen as a software version of a robot (a "softbot") in all these senses.

Initially, the name "intelligent agent" was introduced for programs with a specific architecture, such as the belief–desire–intention (BDI) model that manages the software's perception-planning-action cycles according to given goals.

Gradually, the term was generalized to refer to all types of software that share certain desired properties associated with AI:

Example: Siri

Finally, the agent framework has been used to organize various AI fields into a coherent picture. Other examples: 1, 2