1. Scott Lindner 10/6/04 Pathfinding and Navigational Systems, cont.

2. Outline 2.5: How to get more out of a navigation system Architecture Considerations High-Level Benefits Low-Level Benefits 2.6: Hunting down the player Choose destinations wisely Search “intelligently”

3. Using navigation for more than just route-finding When/where to throw a grenade? Automatically climb a ladder or jump a gap? What locations are useful to hide behind, or lean around? Solution: embed additional information directly into navigational data automatically detect it during a preprocessing step, or encode it by hand

4. Architecture Considerations additional AI layer specializes in movement, steering, animation, and other low-level details that decision maker is not designed for

5. High-Level Point Selection and Pathfinding Any data that AI can easily look at is potentially usable Finding cover and combat tactics –“Bounding Overwatch” –More details later… Throwing grenades –Not practical to calculate trajectories during the game –Encode possible throws directly into the system »must be evaluated from all points to all possible targets

6. Throwing Grenades

7. Agent queries navigation system to find closes allowable point Decision making code will instruct agent how to get to the calculated spot and how to throw his grenade

8. Biasing Edge Cost Normally cost = distance Can make agent avoid a given edge by adding to its cost Stop “cutting corners” SOF2 –Avoid areas where comrades die

9. Biasing Edge Cost ParentPoint = OpenList.GetFromOpenList() For each edge from ParentPoint Switch Edge.type() Case FLY_EDGE: If ( actor.CanFly() ) Edge.cost = actor.FlyBiasCost() OpenList.AddToOpenList(Edge)

10. Low-Level Animation and Steering Path usually handed down as abstract series of points Embedded information will give AI knowledge to perform various behaviors Dynamic movement –Jumping over crevice Eliminates need to script behavior Allows special movement types to be added, all existing combat behavior will automatically make use of it

11. The Navigation Point Provide discrete locations to space where AI agents can move to/from Use spatial volumes as discussed previously Test for reachability and “openness” Sphere (of influence)

12. The Navigation Point

13. Obstacle Visibility Vertical tests (height) Used to find locations to hide behind Run collision tests, until you don’t hit anything Previous C and D Side to side test Near a corner? Run collision tests perpendicular Previous A and B

14. Navigation Edge Points alone not sufficient for navigation Adding additional information into edges is very useful Certain characters might have features making traversal possible/impossible –Size »Maximum size can be calculate by running collision tests and anything <= that can pass through

15. Situations Over a crevice – jump? fly? Navigation points in the air Only applicable to agents who can fly, ignored by others Over an obstacle - vault? Jump? Through a movable obstacle Open a door Move a box Embed all animations directly into respective edge

18. Movable Objects Retry collision test after removing/relocating object If next test succeeds, edge is still valid Any complex sequence of behaviors (reach out, pull door open, step back, etc) all embedded in navigation system

19. Slopes, Stairs, Ladders, etc Slopes tell about surroundings –Vertical = ladder/rope –Otherwise = stairs/slope

20. 2.6: Hunting Down the Player in a Convincing Manner Goal is to have NPC convincingly chase and hunt a player –Spotted by guard, etc. Today’s gamers expect intelligent behavior Problem is AI can cheat, following too direct of a path –Leads to “unfair” results Solution: ensure that the agent explores and looks like it’s trying to find it’s nonvisible target by a process of search, rather than direct path

21. Enabling a Range of Behaviors Must be able to control scope of the search Bumbling/wandering agent Agent who searches with purpose, decisively These allow us to control how quickly discovery occurs and how direct the resulting path is

22. Approach Agent’s objective: get closer to player over time –Utilize pathfinding Requires a given destination –Usually places agent closer to player –Simple idea is to head to players current location »does not produce good gameply (player hiding) Solution: come up with more interesting destinations

23. Application Requirements Pathfinding system is capable of establishing relatively quickly the accessibility of the world Method is iterative, therefore requires multiple calls to give desire behavior Knowledge of hunting character and target player Facility of “line of sight” – can we see the player? Basic approach is to periodically generate new search destination over time destinations actually intermediate destinations

24. Possible Problems? Since process is iterative, could lead to lots of calls being made and many paths being constructed (in theory) Solution: (see requirements) generation of search destinations is a relatively infrequent even in AI terms. Unlikely necessary to calculate a new character search destination more frequently than 1 character / second [McLean02]. Intervals lasting several seconds will be perfectly acceptable for many games

25. 3 Possible Scenarios During each update decide which of the 3 currently apply The player is visible The player was recently seen The player has never been seen

26. Scenario One: The Player is Visible Hunting behavior is over Transition into an appropriate attacking behavior Look for cover within weapon range? Move to that spot? Engage in hand to hand combat?

27. Scenario Two: The Player Was Recently Seen NPC does not currently have line of sight Move towards last seen player location –Leads to “fun” gameplay Trick agent by circling an object Need to maintain “last seen” location Store time and location Important to clear “last seen” when agent begins moving on its route Otherwise agent might fall into states that are inappropriate due to out-of-date info

28. Scenario 3: The Player Has Never Been Seen Must create a search location, rather than extract one from the AI system Randomly generate a location somewhere close to player (within a certain radius) –Cheap to do, but not very purposeful or guided Solution: same idea, but slightly different

29. Scenario 3, cont. Generate two random parameters –Direction heading Direction vector to true player’s location Then alter vector subject to a certain window of variance –Straight line distance to travel Lies within given interval »Consider the interval a multiplier of actual distance to player

30. Scenario 3, cont

31. Keep in mind that since destination is randomly generated, it might not actually be accessible Route to closest location that IS accessible Existing pathfinding system already offers this functionality Also, route to destination wont be straight line, but again, built into pathfinding system Must update vision more frequently than generation of destinations in case we see the player

32. Scenario 3, cont.

33. Putting it all together Going back to previous slide

34. Enabling a Range of Behaviors Must be able to control scope of the search Bumbling, wandering, accidental, discovery of player, search path that is purposeful, efficient, direct These allow us to control how quickly discovery occurs and how direct the resulting path is

35. Putting it all together Purposeful characters that appear to find the player quickly given values of Smin and Smax close to 1.0 with a small Φ. “Bumbling” agents will require values for Φ, Smin, and Smax far from 1.0

36. Extending to 3D Easily extend to 3D by generating points inside 3D cones Cost very insignificant and approach still very successful Relative sizes of regions to be searched will vary non-linearly (i.e. halving the radius of the cone will more than halve the search space) Agent will find character twice as fast

37. Other Considerations Agent must always be reconsidering its current destination As to not ignore a player who runs directly in front of him on his way to a destination Static players/dynamic players »Dynamic means MANY more calls Can respond to other percepts Sounds Other teammates

38. Sources AI Game Programming Wisdom 2 Mclean, Alex, “An Efficient AI Architecture Using Prioritized Task Categories,” AI Game Programming Wisdom, Charles River Media, 2002