Welcome to my home-page

Alfred Schurmann

The build in intelligence of the motivated autonomous robot

There are many papers about robot intelligence. Some ideas of making robot intelligent are rather naive. I consider intelligence of the motivated autonomous robot in other way - as outlined below, I build in some foundation stones of intelligence in the robot.

The following components of the motivated autonomous robot are the basis for its intelligent behavior:
*   a tree like structured knowledge base of object and situation models,
*   emotional patterns associated to said object and situation models,
*   activity schemas which enable the robot to perform complex activities,
*   the motivation of the robot to execute an activity,
*   the perceiving sub-system which focuses its attention on the most important objects and situations (at present).

1. Abstract object and situation models in the knowledge base and its intelligent updating

The structure of the knowledge base is such that each object or situation model is a node in a tree like graph. The successors of an object node are more concrete object models than the predecessor model; they are instantiations of the predecessor object model; e.g. the successor of the object node "apple" are models of different kinds of apples, whereas the "apple" node (it is the model of the class of apples) is a successor of the node/model "fruit" (it is the model of the class of fruits). An object model has all properties of its predecessor model/node. this enables to have very abstract object models in the knowledge base. Abstract object and situation models are a foundation stone of intelligence, because, when the robot perceives an object which it has not seen, it can recognize/identify this object as belonging to a class of objects i.e. as being a case of an abstract object model - thus, in this way intelligent identify before not known objects.

in the case of a new object, which model is not in the knowledge base, the robot can build the model of this object (doing additional activities to identify more features of this object) and insert this model into the appropriate place in the tree knowledge base; in this way the robot intelligent enlarge its knowledge about its surrounding.

In the similar way not known situations can be classified (identified as being a case of an abstract situation model). Thus, when the robot perceives an unknown dangerous situation then it would probably recognize it as belonging to a class of dangerous situations - thus, intelligent recognize the not exactly known current surrounding.

The robot can also verify whether a recognized concrete object has all/or new properties with regard to its model in the knowledge base and update this object model -thus, intelligent reacting of changes of an object.
in said way the robot can adapt intelligently its knowledge base to a new surrounding; e.g. when a household robot changes the flat or a new furniture are in the flat.

The knowledge base is used also for associative perception of object and situations, in the following way: when the robot perceives an object as a visual image, by a program for visual identification, then it searches, in the knowledge base, for the object model which visual pattern best matches with the identified image; it searches from the top along the path where the visual image best matches to the visual patterns in models connected to the nodes. After it found the object model having the best matched visual pattern, it associates the identified image with this object model, thus with all properties of this model. This is a simple solution of the associative problem when robot perceives an object, which before was not solved.

2. Valuation of objects and situations by means of emotional stimulus patterns

To each object and situation model may be associated a stimulus pattern with respect to a need of the robot. This enables to calculate the current stimulus intensity of a perceived object or situation. This stimulus intensity measures how good (if it is positive) or bad (if it is negative) and how important (how great is the absolute value) the perceived object or situation is at present - thus, it can intelligent valuate objects and situations, which enable the robot to ignore not important, at present, objects and situations and consider only the most important ones.

3. Situations, activity schema and the motivation of the robot to execute it

Usually in papers, a robot considers its surrounding as an appearance of objects. This leads to unintelligent behavior; examples: "if tap (faucet) open then close it", if a book on a floor then take it and put on the shelf (also when e.g. children use it)".
A robot must consider its current surrounding as an appearance of situations, otherwise it cannot handle intelligently. The autonomous robot reacts (i.e. does activities) only to situations, as outlined below.

An activity schema of the robot is a description of a complex activity of the robot, e.g. the activity schema for preparing a lunch, the activity schema for cleaning a room. Such activity schema may have control sub-activities which enable the robot to operate and control other machines. An activity schema has initial and goal situations. The robots starts the execution of an activity from one of its initial situation, and this execution should lead to one of the goal situations; if it does not achieve a goal situation then the execution of this activity failed. Thus, the robot does an activity only if it is in an initial situation of this activity.

However, usually the robot is in several situations which are initial situations of some activity schemas. In order to decide which activity it should execute, at present, it uses a procedure, which calculates the current motivation to execute an activity. This motivation procedure is based on said in 2 valuation of stimulus intensities of some situations (especially the initial, goal and other end situations) of the activity schema. By means of this motivation, taking into account the time in which an activity must be done, the robot determines the activity which it is going to execute - it does the activity which execution is most important at present. Thus the robot determines intelligently which activity it currently does. Such decision problem have only full-autonomous robots; not full-autonomous robots are not confronted with such problem.

4. Focusing its attention on the most important objects and situations

When perceiving objects and situations in its current surrounding, the robot focuses its attention and perception on these objects and situations which are most important at present, i.e. which have great negative or positive stimulus intensities (see 2. Valuation of objects...). This is one of the basis for intelligent behavior of the robot, because it handles the most important situations and ignores unimportant ones.

Robot which observes all situations in its surrounding with equal attention cannot observe efficiently actually important situations, which leads to handling an unimportant situation and ignoring an important one - this is a very unintelligent behavior.

The perceiving sub-system of the robot stores/remembers all objects and situations in its actually perceived surrounding, so that, e.g. it knows which objects are in the living room when it has come from it to the kitchen. Such knowledge is necessary to behave intelligently.

5. Understanding and using a very primitive language

We may connect names to object and situation models in the knowledge base (s. "1. Abstract...") in different languages, e.g. to the model of an apple the words "apple" and "Apfel". The meaning of such name/word is the associated object or situation model. For simple clauses, we could build abstract situation patterns. The robot would understand such clauses and could formulate clauses according to said situation patterns. Thus, the robot could understand and use a very primitive language.

However, the understanding of and using a language by the robot is very limited, because (a) the knowledge base contains only models of object and situations with which it is directly confronted, and (b) we do not know the rules and language patterns according to which a human builds sentences - the robot has not such intelligence.

A.Schurmann                           Modified:            September 24, 2006

Goto startsite