Topic 4

Knowledge representation, rule-

based systems
Steering around obstacles
Knowledge Representation
Rule-Based Systems
Synthesising Movement with an RBS

Reading: Champandard Chapters 9 -12

Links to RBS, Knowledge Representation on website

ICT219 1
More on Obstacle Avoidance
• Continuing to develop our obstacle-avoidance code, Champandard uses
the concept of steering behaviours (Reynolds, 1999)

• Assume an animat which can move (using a locomotion layer). Assume

a physics system which has friction and obeys Newtonian laws of
motion

• The velocity of an animat is crucial. Recall that velocity is a vector,

meaning it has both magnitude and direction. A steering force is also a
vector, which when added to the animat’s existing velocity, will change
its speed and direction.

• We continually compute new steering forces to control movement

resulting motion
steering force

existing motion

ICT219 2
Steering, seeking and fleeing
• Velocities and steering forces are both limited at some max value
truncate(steering_force, MAX_STEERING_FORCE) //cap force to max
velocity = velocity + steering_force; // accumulate vectors to get new vel
truncate(velocity, MAX_VELOCITY) //cap animat speed to max
Motion->move(velocity) //use interface to locomotion layer to apply the new vel

• This can be the basis of other behaviours such as seeking, or moving to

a target: compute a steering force that turns an animat toward some
target.
steering force

resulting motion

existing motion

• The desired force is computed from the distance and bearing to the
target and relative to the existing velocity of the animat
• It is well to slow the animat down as it approaches a target because
at maximum velocity it will overshoot and oscillate
• Fleeing behaviour may be computed ICT219by simply negating the seeking3
vector, so that the direction of travel is away from the target
A Basic Obstacle-Avoidance Algorithm
function avoid_obstacles2
check the Vision senses for free space in front, left and right
if a frontal collision is projected (using Tracewalk)
find the furthest obstacle on left or right
determine the best side to turn towards
compute the desired turn to seek the free space
endif

if the front obstacle is within a threshold distance

proportionally determine a breaking force (slow down or stop)
endif

if there is an obstacle on the left side

proportionally adjust the steering force to step right
endif

if there is an obstacle on the right side

proportionally adjust the steering force to step left
endif

apply steering and breaking forces

end function avoid_obstacles2 ICT219 4

Knowledge Representation - Basics
• In AI (and human) problem-solving, it matters a great deal how a
problem is represented. Knowledge representation is the question of
how human knowledge can be encoded into a form that can be handled
by computer algorithms and heuristics
• Knowledge representation languages have been developed to try to
make representations as expressive, consistent, complete and
extensible as possible

• An ideal representation would

- involve a good theory of intelligent reasoning
- provide symbols that served as adequate surrogates for objects,
events etc in the world
- make ontological commitments (about what objects exist, and how to
classify them)
- provide data structures which can be efficiently computed
- provide data structures which human beings can express all kinds
ICT219 5
of knowledge
Knowledge Representation - Basics
• We have already discussed one KR issue – whether knowledge should
be encoded declaratively or procedurally (almost certainly need both)

• The notation of a KR is the symbols and syntax – what is written down

• The denotation is the semantics, what the symbols refer to, or what you
can do with them

• In most KR, knowledge is a mixture of explicitly represented notation,

and implicit knowledge available via inference

• There are a lot of unanswered questions about how to gather, organise,

store and use knowledge-bearing data structures

• There is also a lot of research devoted to the processes (algorithms or

heuristics) by which knowledge-bearing data structures can be
manipulated

ICT219 6
Formalisms
• Symbols – we already use representations in basic programming
left_obstacle_distance = 4.0; // distance to left obstacle is 4 units
left_obstacle_colour = ‘’blue’’; // colour blob there is in the blue
left_obstacle_identity = ‘’unknown’’; // nothing from the recogniser though
right_obstacle_distance = 12.0; // distance to left obstacle is 4 units

• Object-attribute-value To avoid the need to create many objects,

attributes are represented as functions with two parameters: A(o,v)

distance (left_obstacle, 4.0);

distance (right_obstacle,12.0);

colour (left_obstacle, blue);

• More natural since most attributes would be fetched from the world by a
function ICT219 7
Formalisms
• Frames – are a labelled package of slots, each with a value. The values
might be pointers to other frames, or actors, which may launch
procedural code
frame_left_obstacle:
distance: (4.0)
colour: (blue)
identity: (frame_fountain)
on_proximity: <avoid_right 50.0f>

• Semantic Networks – graphs with nodes representing entities linked by

arbitrary relationships
Living_Thing Inanimate
is-a is-a

Bird Container
is-a is-a

Yellow Tweety Perch Cage Door

colour supported-by has-part has-part

ICT219 8
Formalisms
• Conceptual Graphs – more compact and flexible than semantic
networks. Relationships get the status of nodes, and is-a links are made
implicit

GIRL:Sue AGNT EAT MANR FAST

OBJ PIE

• The graphs are suited to the expression of commonsense notions. The

above graph represents the deep meaning underlying the sentence
“Sue is quickly eating a pie.”

• Conceptual graphs can also have actors, like frames

• Such graphs can be manipulated according to canonical truth-

preserving algorithms. These operations allow reasoning to carried out
in terms of the graphs
ICT219 9
Formalisms
• Formalisms
Rules – antecedent clause (condition) related to a consequent clause
(action) by implication
if (A and B) THEN S1

• Antecedent clause is a boolean expression (possibly containing AND

OR and NOT). Consequent clause can be any assertion or function call
• Fuzzy rules – rules containing fuzzy logic terms (see later)

• Subsymbolic patterns - eg networks with numerically weighted links

eg. Bayesian
network

• None! eg sensori-motor behaviours reacting directly to the world

ICT219 10
 Knowledge Engineering
• Specification procedure – effort to pin down a rather scruffy kind of
knowledge engineering suitable for novelle game AI

• Sketching – informally draft some ideas about how to encode the

important data structures and methods. Input (which inputs are
available, which are needed, in what form?), output (what primitive
motor actions are needed?) and context (what goes on behind the
interfaces, what variables are involved, how can things be simplified?).

• Formalising – the rough design is refined towards a formal specification

for programming ie properly documented

• Rationalising - The representations are tested for consistency, so that

the parts will interoperate, with each other and with existing machinery

• This processes is very creative and may loop back (like the Software
Development Life Cycle)
ICT219 11
Rule Based Systems - Basics
• This is a real success story of AI – tens of thousands of working
systems deployed into many aspects of life

• Terms: a knowledge-based systems are really anything that stores and

uses explicit knowledge (at first, only RBSs, later other kinds of system);
a rule-base system (or production system) is a KBS in which the
knowledge is stored as rules; an expert system is a RBSs in which the
rules come from human experts in a particular domain

• Can be used for problem-solving, or for control (eg of animat motion)

• In an RBS, the knowledge is separated from the AI reasoning

processes, which means that new RBSs are easy to create

• An RBS can be fast, flexible, and easy to expand

ICT219 12
RBS Components
• Working memory – a small allocation of memory into which only
appropriate rules are copied

• Rulebase – the rules themselves, possibly stored specially to facilitate

efficient access to the antecedent

• Interpreter – The processing engine which carries out reasoning on the

rules and derives an answer

• An expert system will likely also have an Explanation Facility - keeps

track of the chain of reasoning leading to the current answer. This can
be queried for an explanation of how the answer was deduced.

• Thus unlike many other problem-solving AI modules, an expert system

does not have to be a “black box” ie it can be transparent about how it
arrives at answers

ICT219 13
Rule Based Systems - Architecture

Keeps a trace of chain of rules

for the current solution
optionally
EXPLANATION
FACILITY

ICT219 14
 Rule Based Systems – Forward Chaining
• To reason by forward chaining, the interpreter follows a recognise-act
cycle which begins from a set of initial assertions (input states set in the
working memory) and repeatedly applies rules until it arrives at a
solution (data driven processing)

• Match – identify all rules whose antecedent clauses match the initial
assertions
• Conflict resolution – if more than one rule matches, choose one
• Execution – the consequent clause of the chosen rule is applied, usually
updating the state of working memory,
ICT219 finally generating output 15
 Rule Based Systems - Backward Chaining
• In backward chaining, the interpreter begins from a known goal state
and tries to find a logically supported path back to one of the initial
assertions (goal-directed inference)

en D

• Match – identify all rules whose consequent clauses match the working
memory state (initially the hypothesised goal state)
• Conflict resolution – if more than one rule matches, choose one
• Execution – the state of working memory is updated to reflect the
antecedent clause of the matched rule
ICT219 16
Rule Based Systems – Conflict Resolution
• The interpreter must choose one path to examine at a time, and so
needs a method of conflict resolution in case of multiple rules
• Methods include:

- first come, gets served: choose the first rule that applies
- rules are ordered into priorities (by expert): choose the highest rank
- most specific rule is chosen: eg, apply the rule with most elements in
antecedent
- choose a rule at random, and use that
- keep a historical trace, and allow this to chose a different rule next
time

• If the first chosen rule leads to a dead end, it should be possible to try
another, provided a trace is kept

• Note that this is another example of search through a graph of

possibilities ICT219 17
Rule Based Systems - Development

Explanation
Facility

• The process of encoding human knowledge into an expert system is

called knowledge elicitation
• It can be quite difficult to elicit expert knowledge from experts
- they might not know how they know
- they are very busy people – can’t get their attention for long
- some knowledge might be difficult to codify as rules
- different experts might disagree on facts and methods
- facts and methods may change => need to maintain the system
ICT219 18
• Programmer may customise an expert system shell
Synthesising Movement with an RBS
• Champandard describes a RBS system which implements wall-
following behaviour:
1. If no wall is present and the animat was not already following one,
move forward by a random amount
2. If a wall ahead is detected, the animat should turn away, neglecting any
following of a wall at the side
3. If there is a wall to the side and not at the front, it should be followed by
moving forward
4. If no wall is present and one was being followed, the animat should turn
towards the side where the wall was last detected

• 1 and 4 are similar, but depend on what the animat has been doing,
thus a purely reactive agent cannot generate this behaviour

• However, an RBS can; just use an internal symbol to remember what

the animat has been doing

• In a control forward chaining used rather than backward chaining

(why?) ICT219 19
How to Declare the Rulebase
• The contents of the working memory and the rulebase is defined in XML
<memory>

} working memory, set values, defaults

<Symbol name="sideWall" />
<Symbol name="turnTowards" value="false" />
......
......
</memory>

<rulebase>
<Rule> default values are called ‘ground’
<conditions>
<Symbol name="following" value="false" ground=“true”/>

} rule base, condition-action pairs

<Symbol name="frontWall" value="false" />
<Symbol name="sideWall" value="false" />
</conditions>
<actions>
<Symbol name="following" value="false" />
</actions> conjunction (AND) between clauses
</Rule> is implicit
........
........
... other rules... more than one action is allowed
........
</rulebase>

ICT219 20
How to Use the Rulebase Interface
• Native symbols must be synchronised so as to relay input from sensors
into the RBS system at runtime

• Register a variable declared in the working memory XML by

SetSensor( &SensorSideWall, ‘’sideWall’’); //&Sensor.. is pointer to C++ var

• This variable is now synchronised with the one in the working memory

• We can use Get and Set accessor methods to fetch and change the
values in working memory:
void Set (const string& symbol, const bool value ); //sets a native var

bool Get (const string& symbol) const;// returns boolean state of the var

• Actions are implemented as functors (classes that store functions).

Each action can then be a method in the same class. The C++ native
variables can also be included for convenience.

• You can even add rules dynamically – not recommended to begin with!
ICT219 21
How the RBS Simulator Works
• FEAR has a RBS simulator in the ‘modules’ directory
• The working memory and rulebase set up with appropriate XML
• Needs native functions linked to sensors and effectors
• Interpreter (used for animat control) cycles through these steps:
1. Update working memory variables from inputs using CheckSensors();
2. Scans rules (top to bottom), looking for a match using ApplyRules().
Rules that do not match are skipped. If all rules match, the defaults are
applied, then overridden by rule specifications. If no matches are found,
the default values are applied.
3. Effector symbols are scanned by CheckEffectors(), and native function
calls are made if any are true.
• Calling the function Tick() executes one such cycle
• Once the system has been set up, most of the modifications can be
done by just modifying the rules (provided they don’t need new
sensors and effectors)

ICT219 22
Back into Quake 2
• The Stalker demonstration animat uses these methods to implement
wall-following

• Study this carefully in the laboratory this week, using Chapter 12 as

your guide

ICT219 23