What we have covered since the midterm exam.

Chapters 7, 8, 9, 10, additional material

Chapter 7
Raven game
4 types of weapons; what happens when a projectile hits a robot.
triggers; examples; giver triggers
steering behaviors: seek, arrive, wall avoidance, separation
perception, sensor omnipotence, sensory nescience
need for short-term memory
selecting target; how regulators allow different components update at
a slower than the game update rate.
update method

Chapter 8
Navgraphs, course grained graphs, fine grained graphs, flood fill
algorithms, line of sight; cell-space method
path-finding, uses both A* and Dijkstra, edge annotations
general approach to get from A to B when A and B are not on navgraph
general approach to find nearest target of given type
path smoothing
reducing cpu overhead; shortest path lookup table
time-sliced path planning; CycleOnce
hierarchical planning
sticky situations; detecting

Chapter 9
goal-driven behavior; atomic goals, composite goals, goal stack
examples atomic: Wander, TraverseEdge, SeekToPosition,
DodgeSideToSide
examples composite: FollowPath, MoveToPosition, AttackTarget,
Goal_NegotiateDoor
goal arbitration; Goal_Think
six strategy goals
computing desirability
use of random numbers to give robots "personality"
use a queue rather than a stack to process commands as goals

Chapter 10
Fuzzy Logic; set membership, crisp sets, degree of membership
common shapes of membership function, fuzzification
fuzzy definition of AND, OR, NOT
fuzzy linguistic variables
fuzzy rules, how applied
defuzzification; mean of maximum (MOM), centroid, average of maxima
Combs' method, what problem it avoids

Additional material

Behavior trees
Conditions and Actions
composite tasks: Selection, Sequence, RandomSequence, RandomSelector
decorators: Limit, UntilFail
Parallel, libraries of behavior trees

Goal oriented Behavior
Characters have goals that psychologists call drives
Strength of a goal is called insistence
Choose action that reduces strongest drive the most
Multiple drives can be reduced at same time; now are actions
selected to reduce multiple drives the most?

Rule based systems
global database, condition-action rules
cycle: match new facts, resolve conflict set, execute one rule
rules are refractory, modular, natural
RETE network
rule priority
simple version of a rule based system (if-elseif-... that is
repeatedly executed)
monkey gets banana examples

Goal Oriented Action Planning
Find sequence of up to maxdepth actions that produced best future
world (lowest discontentment), do first action in that sequence
how that procedure can be time sliced

IDA* algorithm
Iterative deepening A*
A blend between procedure for goal oriented action planning and A*
Like goap in that it does depth-first search of a tree of possible
future worlds to a maximum depth of maxdepth; like A* in that it
estimates cost of final path through a world and stops increasing
a path when goal world is reached. Explores each path until estimated
cost exceeds cutoff limit.  Repeats this with larger and larger cutoff
value until best path is found.

Tactical and Strategic AI
tactical locations such as cover, shadow, sniping, exposed
Monte Carlo method of estimating fuzzy membership of waypoint
    into type of tactical location
automatically generating waypoints
condensation algorithm for reducing number of waypoints
influence maps
map flooding
Frag maps
convolution
Coordinated action
top-down approach
bottom-up approach
military tactics