More on the AI System

• The perception system.
• We can add "senses" to an actor
  • This is done via a AIPerception component.
  • These can be customized to recognize a set of "senses"
  • Players with this have a OnTargetPerceptionUpdated event
    • Triggered when a stimuli is perceived.
    • Triggered when a stimuli stops.
• We can also add "Stimuli" to an actor
  • This causes the item to generate "stimulus"
  • This is done via a AIPerceptionStimuliSource component.
• He demos how to configure these.
• And how to debug these.
• In the end, he sets up a blueprint on the AI character
  • This registers an actor that has been perceived.
    • IE the AI character "remembers who it saw"
  • And removes this actor a bit after it is not longer perceived.
  • 
• Behavior Trees
  • First introduced about 2004
  • Still a topic of research
  • They represent a way to implement complex agent behaviors.
  • Some tree terminology
    • Trees are represented by nodes (or points in the tree)
    • And edges connecting these nodes.
    • The edges are directed.
    • The node at the top of the tree is known as the root
    • The nodes at the bottom of the tree are known as leaves.
    • There are no cycles or loops
      • Every node but the root had exactly one parent.
      • But a parent may have many children.
  • In behavior trees, control flows
    • From top (the root) to bottom (the leaves)
    • From Left to right among equal children.
  • In general, nodes in a tree represent a task
  • They can be
    • simple, just a single task
    • composite: the combination of several tasks.
  • Task Status
    • A task can be
      • Running
      • Successful
      • Failure
  • Composite nodes are
    • Selectors
      • It will execute the nodes below it.
      • It is considered successful if any of the children nodes are successful.
      • It will fail of all nodes fail.
      • Example: eat dinner
        • Look for something pre-cooked in the kitchen
        • Try to cook something
        • Order something delivered.
        • Go to a restaurant.
      • By placing the nodes in order (left to right) we prioritize which task to attempt first.
      • Again, this succeeds if any one task succeeds.
      • And it only does tasks until one is successful.
    • Sequencers
      • This will execute the nodes below it.
      • For this to be successful all nodes must succeed.
      • Example: Prepare dinner
        • Find a recipe
        • Get the ingredients
        • Follow each step of the recipe
        • Place the food on the table.
      • Again the steps are in order.
      • If any step fails, the entire operation fails.
  • There is a third composite node, simple parallel.
    • These have two sub nodes
    • The first is a simple task.
    • The other can be a complete b tree
    • It will attempt to execute both.
    • Document example
      • Attack an enemy while moving towards a goal
    • The first task is primary.
    • The second task is secondary.
  • Together these concepts lead to an incredibly powerful system.
  • Nodes can have one or more decorators
    • These use blackboard variables to determine if the node should be executed.
• Practical Behavior Trees
  • These consist of two components.
  • The behavior tree.
  • A blackboard.
    • Sort of what the name implies
    • This is a global storage space for the behavior tree.
    • Think of it as working memory.
• In the demo he has you build a task (leaf node)
  • BBT_FindNaviagble Location
  • 
  • Then jumps to a completed tree
  • 
  • The intermediate steps are
    • Return to the Behavior Tree.
    • Right click and select BTT_FindNavigableLocation
    • 
    • Set the TargetLocationKey to be the TargetLocation
    • 
      • This is how you link the individual values on the blackboard to the tasks in the tree.
    • The root will only all composite nodes, and we need to do three tasks so we will use a sequence.
      • Drag off of the box at the bottom of the root.
      • Select Sequence
      • 
    • Add a MoveTo task
      • Set the Blackboard Key to be Target Location
      • 
    • Finally, add a wait node.
• Communicating to a blackboard
  • He does this in two stages
    • He builds communication functions in the controller.
    • These take parameters and set values in the blackboard.
    • This is done via a "key" or the variable name in the blackboard
    • I assume they implement the blackboard at least partially as a hash table (Map, Dictionary)
    • As such, the key name must match the backboard entry completely.
    • 
    • Then he calls these functions from the other code.