Artificial Intelligence  
Lecture 06 Automated Planning  
Edirlei Soares de Lima  
<edirlei.lima@universidadeeuropeia.pt>  
Game AI Model  
Pathfinding  
Steering behaviours  
Finite state machines  
Automated planning  
Behaviour trees  
Randomness  
Sensor systems  
Machine learning  
Decision Making  
In game AI, decision making is the ability  
of a character/agent to decide what to  
do.  
The agent processes a set of information  
that it uses to generate an action that it  
wants to carry out.  
Input: agent’s knowledge about the world;  
Output: an action request;  
Decision Making  
The knowledge can be broken down into external and  
internal knowledge.  
External knowledge: information about the game environment (e.g.  
characters’ positions, level layout, noise direction).  
Internal knowledge: information about the character’s internal state  
(e.g. health, goals, last actions).  
Goal-Oriented Behavior  
So far we have focused on reactive agents: a set of inputs is  
provided to the character, and an appropriate action is  
selected.  
Goal-oriented behavior is an alternative approach. It adds character  
goals/desires to the decision-making process.  
To allow an NPC to properly anticipate the effects and take  
advantage of sequences of actions, a planning process is  
required.  
Automated Planning Techniques.  
Automated Planning  
Planning is the task of finding a sequence of actions (a plan) to  
achieve a goal.  
Example:  
Goal: have(sword)Λ at(castle)  
Plan: go(dungeon), kill(enemy), get(key), go(forest),  
open(chest, key), get(sword), go(castle).  
Plan-based agent process:  
1
2
3
) Formulate a goal;  
) Find a plan;  
) Execute the plan;  
Automated Planning  
A planning problem is usually represented through a planning  
language, such as the PDDL (Planning Domain Definition  
Language).  
PDDL was derived from the original STRIPS model, which is slightly  
more restrictive.  
Planning problem elements:  
Initial State;  
Actions (with preconditions and effects);  
Goal;  
Planning Problem  
Each state is represented as a conjunction of predicates.  
Example: At(Truck1, Melbourne)At(Truck2, Sydney).  
Closed-world assumption: any predicates that are not mentioned are  
false.  
Actions are described by a set of action schemas with  
parameters, preconditions, and effects.  
Example:  
Action(  
Fly(p, f, t),  
PRECOND: At(p, f) Plane(p) Airport(f) Airport(t)  
EFFECT: ¬At(p, f) At(p, t)  
)
Planning Problem  
The precondition defines the states in which the action can be  
executed.  
Example:  
Action(  
Fly(p, f, t),  
PRECOND: At(p, f) Plane(p) Airport(f) Airport(t)  
EFFECT: ¬At(p, f) At(p, t)  
)
Initial State: At(C1, SFO) At(C2, JFK) At(P1, SFO) At(P2, JFK) Cargo(C1) ∧  
Cargo(C2) Plane(P1) Plane(P2) Airport (JFK) Airport (SFO)  
The Fly action can be instantiated as Fly(P1, SFO, JFK) or as Fly(P2, JFK, SFO).  
Planning Problem  
The effect defines the result of executing the action.  
Example:  
Action(  
Fly(p, f, t),  
PRECOND: At(p, f) Plane(p) Airport(f) Airport(t)  
EFFECT: ¬At(p, f) At(p, t)  
)
Initial State: At(C1, SFO) At(C2, JFK) At(P1, SFO) At(P2, JFK) Cargo(C1) ∧  
Cargo(C2) Plane(P1) Plane(P2) Airport (JFK) Airport (SFO)  
Negative predicates are removed from the resulting state (e.g. ¬At(p, f));  
Positive predicates are added to the resulting state (e.g. At(p, t));  
Example Air Cargo Transport  
Init(At(C1, SFO) At(C2, JFK) At(P1, SFO) At(P2, JFK) ∧  
Cargo(C1) Cargo(C2) Plane(P1) Plane(P2) ∧  
Airport (JFK) Airport (SFO))  
Goal(At(C1, JFK) At(C2, SFO))  
Action(  
Load(c, p, a),  
PRECOND: At(c, a) At(p, a) Cargo(c) Plane(p) Airport(a)  
EFFECT: ¬At(c, a) In(c, p)  
)
Action(  
Unload(c, p, a),  
PRECOND: In(c, p) At(p, a) Cargo(c) Plane(p) Airport(a)  
EFFECT: At(c, a) ¬In(c, p)  
)
Action(  
Fly(p, f, t),  
PRECOND: At(p, f) Plane(p) Airport(f) Airport(t)  
EFFECT: ¬At(p, f) At(p, t)  
)
Example Blocks World  
Init(On(A, Table) On(B, Table) On(C, A) ∧  
Block(A) Block(B) Block(C) Clear(B) ∧  
Clear(C))  
Goal(On(A,B) On(B,C))  
Action(  
Move(b, x, y),  
PRECOND: On(b, x) Clear(b) Clear(y) ∧  
Block(b) Block(y) (b x) ∧  
(
b ≠ y) (x y),  
EFFECT: On(b, y) Clear(x) ¬On(b, x) ∧  
Clear(y)  
¬
)
Action(  
MoveToTable(b, x),  
PRECOND: On(b, x) Clear(b) Block(b) ∧  
(
b ≠ x),  
EFFECT: On(b, Table) Clear(x) ¬On(b, x)  
)
Planning Algorithms  
The description of a planning problem defines a search  
problem: we can search from the initial state looking for a goal.  
Planning approaches:  
Progressive: forward state-space search;  
Regressive: backward relevant-states search;  
Forward State-Space Search  
take c3  
take c2  
move r1  
Backward Relevant-States Search  
g1  
a1  
a4  
g4  
g2  
g0  
a2  
a3  
s0  
a5  
g5  
g3  
Planning Domain Definition Language  
A planning problem is usually represented through a planning  
language, such as the PDDL (Planning Domain Definition  
Language).  
PDDL was derived from the original STRIPS model, which is slightly  
more restrictive.  
Planning problems specified in PDDL are defined in two files:  
Domain File: types, predicates, and actions.  
Problem File: objects, initial state, and goal.  
PDDL Syntax  
Domain File:  
(define (domain <domain name>)  
(
(
(
<
<
[
<
:requirements :strips :equality :typing)  
:types <list of types>)  
:constants <list of constants>)  
PDDL code for predicates>  
PDDL code for first action>  
...]  
PDDL code for last action>  
)
Problem File:  
(define (problem <problem name>)  
(
<
<
<
:domain <domain name>)  
PDDL code for objects>  
PDDL code for initial state>  
PDDL code for goal specification>  
)
PDDL Example Problem  
“There is robot that can move between two rooms  
and pickup/putdown boxes with two arms. Initially,  
the robot and 4 boxes are at room 1. The robot must  
take all boxes to room 2.”  
Room 1  
Room 2  
PDDL Domain File  
Types:  
(:types room box arm)  
Constants:  
(:constants left right - arm)  
Predicates:  
robot-at(x) true if the robot is at room x;  
box-at(x, y) true if the box x is at room y;  
free(x) true if the arm x is not holding a box;  
carry(x, y) true if the arm x is holding a box y;  
(:predicates  
(
(
(
(
robot-at ?x - room)  
box-at ?x - box ?y - room)  
free ?x - arm)  
carry ?x box ?y - arm)  
)
PDDL Domain File  
Action: move the robot from room x to room y.  
Precondition: robot-at(x) must be true.  
Effect: robot-at(y) becomes true and robot-at(x) becomes  
false.  
(:action move  
:
:
:
parameters (?x ?y - room)  
precondition (robot-at ?x)  
effect (and (robot-at ?y) (not (robot-at ?x)))  
)
PDDL Domain File  
Pickup Action:  
(:action pickup  
:
:
parameters (?x - box ?y - arm ?w - room)  
precondition (and (free ?y) (robot-at ?w)  
(
box-at ?x ?w))  
effect (and (carry ?x ?y) (not (box-at ?x ?w))  
not(free ?y)))  
:
(
)
Putdown Action:  
(:action putdown  
:
:
:
parameters (?x - box ?y -arm ?w - room)  
precondition (and (carry ?x ?y) (robot-at ?w))  
effect (and (not(carry ?x ?y)) (box-at ?x ?w)  
(free ?y))  
)
PDDL Domain File  
define (domain robot)  
(
(
(
(
(
:requirements :strips :equality :typing)  
:types room box arm)  
:constants left right - arm)  
:predicates  
(
(
(
(
robot-at ?x - room)  
box-at ?x - box ?y - room)  
free ?x - arm)  
carry ?x - box ?y - arm)  
)
(
:action move  
:
:
:
parameters (?x ?y - room)  
precondition (robot-at ?x)  
effect (and (robot-at ?y) (not (robot-at ?x)))  
)
(
:action pickup  
:
:
:
parameters (?x - box ?y - arm ?w - room)  
precondition (and (free ?y) (robot-at ?w) (box-at ?x ?w))  
effect (and (carry ?x ?y) (not (box-at ?x ?w)) (not(free ?y)))  
)
(
:action putdown  
:
:
:
parameters (?x - box ?y -arm ?w - room)  
precondition (and (carry ?x ?y) (robot-at ?w))  
effect (and (not(carry ?x ?y)) (box-at ?x ?w) (free ?y))  
)
)
PDDL Problem File  
Objects: rooms, boxes, and arms.  
(:objects  
room1 room2 - room  
box1 box2 box3 box4 - box  
left right - arm  
)
Initial State: the robot and all boxes are at room 1.  
(
)
:init  
(
(
(
(
(
(
(
robot-at room1)  
box-at box1 room1)  
box-at box2 room1)  
box-at box3 room1)  
box-at box4 room1)  
free left)  
free right)  
PDDL Problem File  
Goal: all boxes must be at room 2.  
(
:goal  
(
and (box-at box1 room2)  
(box-at box2 room2)  
(box-at box3 room2)  
(box-at box4 room2)  
)
)
PDDL Problem File  
(
(
define (problem robot1)  
:domain robot)  
(:objects  
room1 room2 - room  
box1 box2 box3 box4 - box  
left right - arm  
)
(
:init  
(
(
(
(
(
(
(
robot-at room1)  
box-at box1 room1)  
box-at box2 room1)  
box-at box3 room1)  
box-at box4 room1)  
free left)  
free right)  
)
(
:goal  
(and  
(
(
(
(
box-at box1 room2)  
box-at box2 room2)  
box-at box3 room2)  
box-at box4 room2)  
)
)
)
PDDL Planners  
HSP Planner - https://github.com/bonetblai/hsp-planners  
Heuristic Search Planner;  
Compiled version for windows (cygwin):  
http://edirlei.com/aulas/ia_2013_1/HSP-Planner.zip  
Online PDDL Planner:  
Editor: http://editor.planning.domains/  
Remote API: http://solver.planning.domains/  
HSP Planner  
Executing the planner:  
hsp.exe robot-problem.pddl robot-domain.pddl  
Extra parameters:  
Search direction: -d backward ou forward  
Search algorithm: -a bfs ou gbfs  
HSP Planner  
Forward search:  
Backward search:  
(
(
(
(
(
(
(
(
(
(
(
(
(
PICKUP BOX1 LEFT ROOM1)  
MOVE ROOM1 ROOM2)  
PUTDOWN BOX1 LEFT ROOM2)  
MOVE ROOM2 ROOM1)  
PICKUP BOX2 LEFT ROOM1)  
MOVE ROOM1 ROOM2)  
PUTDOWN BOX2 LEFT ROOM2)  
MOVE ROOM2 ROOM1)  
PICKUP BOX3 LEFT ROOM1)  
PICKUP BOX4 RIGHT ROOM1)  
MOVE ROOM1 ROOM2)  
(PICKUP BOX4 RIGHT ROOM1)  
(PICKUP BOX3 LEFT ROOM1)  
(MOVE ROOM1 ROOM2)  
(PUTDOWN BOX4 RIGHT ROOM2)  
(PUTDOWN BOX3 LEFT ROOM2)  
(MOVE ROOM2 ROOM1)  
(PICKUP BOX2 RIGHT ROOM1)  
(PICKUP BOX1 LEFT ROOM1)  
(MOVE ROOM1 ROOM2)  
(PUTDOWN BOX2 RIGHT ROOM2)  
(PUTDOWN BOX1 LEFT ROOM2)  
PUTDOWN BOX3 LEFT ROOM2)  
PUTDOWN BOX4 RIGHT ROOM2)  
Online PDDL Planner  
Online PDDL Planner  
Resulting plan:  
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
pickup box1 left room1)  
move room1 room2)  
putdown box1 left room2)  
move room2 room1)  
pickup box2 left room1)  
move room1 room2)  
putdown box2 left room2)  
move room2 room1)  
pickup box3 left room1)  
move room1 room2)  
putdown box3 left room2)  
move room2 room1)  
pickup box4 left room1)  
move room1 room2)  
putdown box4 left room2)  
PDDL Simple Game Situation  
The objective of the NPC is to kill the player, but he can't do  
much without a weapon.”  
The game world comprises three places: store, street and a house;  
There is a gun at the store;  
The NPC is at the street;  
The player is at the house;  
Gun  
NPC  
Player  
House  
Store  
Street  
PDDL Simple Game Situation  
(define (domain simplegame)  
(:requirements :strips :equality :typing)  
(:types location character enemy weapon)  
(:predicates  
(
(
(
(
at ?c ?l)  
path ?l1 ?l2)  
has ?c ?w)  
dead ?c)  
)
(:action go  
:parameters (?c - character ?l1 - location ?l2 - location)  
:precondition (and (at ?c ?l1) (path ?l1 ?l2))  
:effect (and (at ?c ?l2) (not (at ?c ?l1)))  
)
(:action get  
:
:
:
parameters (?c - character ?w - weapon ?l - location)  
precondition (and (at ?c ?l) (at ?w ?l))  
effect (and (has ?c ?w) (not (at ?w ?l)))  
)
(:action kill  
:parameters (?c - character ?e - enemy ?w - weapon ?l - location)  
:precondition (and (at ?c ?l) (at ?e ?l) (has ?c ?w))  
:effect (and (dead ?e) (not(at ?e ?l)))  
)
)
PDDL Simple Game Situation  
(
define (problem npc1)  
(:domain simplegame)  
(:objects  
store street house - location  
npc - character  
player - enemy  
gun - weapon  
)
(:init  
(
(
(
(
(
(
(
at npc street)  
at player house)  
at gun store)  
path store street)  
path street store)  
path street house)  
path house street)  
)
(:goal  
(
and  
dead player)  
(
)
)
)
Exercise 1  
1
) Implement the PDDL domain and problem files to solve the following  
problem: “A giant dragon is attacking the castle and John must find a  
way to kill the dragon!”  
strong weapon (magic bow)  
weak weapon (sword)  
inside  
John  
weak troll  
Store  
chest key  
River  
chest  
strong dragon  
Castle  
Town  
Forest  
Cave  
Additional information:  
John can not leave a location if there is a alive enemy there;  
The weak troll can be killed with the weak weapon (sword);  
The chest is closed. It can be opened with the chest key;  
There is a strong weapon inside of the chest (magic bow);  
The dragon can only be killed with a strong weapon (the magic bow);  
Automated Planning in Unity  
The best way to add automated planning to a Unity project is  
by implementing the planning algorithm directly in Unity.  
Starting point: C# PDDL Parser - https://github.com/sunsided/pddl  
Alternatively, we can use a modified version of the HSP  
Planner (written in C) as a standard alone application that can  
be executed by an Unity script to generate plans.  
http://edirlei.com/aulas/game-ai/HPS-Planner-Unity.zip  
Not an efficient solution. Use it only for prototyping purposes.  
Another option: use the online planning service API:  
http://solver.planning.domains/  
Limitations: internet connection, speed, server overload, ...  
Automated Planning in Unity  
Executing the HSP Planner in Unity:  
using System.Diagnostics;  
..  
try{  
Process plannerProcess = new Process();  
Relative path of the HSP  
exe in the project folder.  
.
plannerProcess.StartInfo.FileName = "Planner/hsp2.exe";  
plannerProcess.StartInfo.CreateNoWindow = true;  
plannerProcess.StartInfo.Arguments = "Planner/game-problem.pddl  
Planner/game-domain.pddl";  
plannerProcess.StartInfo.UseShellExecute = false;  
plannerProcess.StartInfo.RedirectStandardOutput = true;  
plannerProcess.Start();  
plannerProcess.WaitForExit();  
while (!plannerProcess.StandardOutput.EndOfStream){  
UnityEngine.Debug.Log(plannerProcess.StandardOutput.ReadLine());  
}
}
catch (System.Exception e){  
UnityEngine.Debug.Log(e);  
Processes the plan actions  
individually.  
}
Automated Planning in Unity - Example  
Simple Game Situation Example: The objective of the NPC is  
to kill the player, but he can't do much without a weapon.”  
Player  
House  
Street  
Gun  
NPC  
Store  
public class PlanAction {  
public string name;  
public List<string> parameters;  
public Status status;  
Class to store and interpret planner  
actions.  
public PlanAction(string action){  
string temp = "";  
name = "";  
parameters = new List<string>();  
public enum Status { Ready,  
Executing,  
Completed  
};  
foreach (char c in action){  
if (c == ' '){  
if (name.Equals(""))  
name = temp;  
else  
parameters.Add(temp);  
temp = "";  
}
else if (c == ')')  
parameters.Add(temp);  
else if (c != '(')  
temp += c;  
}
status = Status.Ready;  
}
}
public class NPCPlanner : MonoBehaviour {  
private List<PlanAction> plan;  
private int currentAction;  
private NavMeshAgent agent;  
public WaypointInfo[] waypoints;  
void Start(){  
plan = new List<PlanAction>();  
agent = GetComponent<NavMeshAgent>();  
currentAction = 0;  
[
public struct WaypointInfo  
{
System.Serializable]  
public string name;  
public Transform waypoint;  
}
try{  
Process planner = new Process();  
planner.StartInfo.FileName = "Planner/hsp2.exe";  
planner.StartInfo.CreateNoWindow = true;  
planner.StartInfo.Arguments = "Planner/game-problem.pddl  
Planner/game-domain.pddl";  
planner.StartInfo.UseShellExecute = false;  
planner.StartInfo.RedirectStandardOutput = true;  
planner.Start();  
planner.WaitForExit();  
while (!planner.StandardOutput.EndOfStream){  
plan.Add(new PlanAction(planner.StandardOutput.ReadLine()));  
}
}catch (System.Exception e){  
UnityEngine.Debug.Log(e);  
}
}
void Update(){  
if (currentAction < plan.Count){  
if (plan[currentAction].status == Status.Ready){  
DoAction(plan[currentAction]);  
}
if (plan[currentAction].status == Status.Executing){  
CheckAction(plan[currentAction]);  
}
if (plan[currentAction].status == Status.Completed){  
currentAction++;  
}
}
Just an example. Usually you  
should play an animation.  
}
void DoAction(PlanAction action){  
if (action.name.Equals("GO")){  
agent.destination = GetWaypoint(action.parameters[2]);  
action.status = Status.Executing;  
}
else if (action.name.Equals("GET")){  
Destroy(GameObject.FindGameObjectWithTag(action.parameters[1]));  
action.status = Status.Executing;  
}
else if (action.name.Equals("KILL")){  
Destroy(GameObject.FindGameObjectWithTag(action.parameters[1]));  
action.status = Status.Executing;  
}
}
void CheckAction(PlanAction action){  
if (action.name.Equals("GO")){  
if (IsAtDestionation())  
action.status = Status.Completed;  
}
else if (action.name.Equals("GET")){  
action.status = Status.Completed;  
Usually you need to wait  
until the animation ends.  
}
else if (action.name.Equals("KILL")){  
action.status = Status.Completed;  
}
}
Vector3 GetWaypoint(string name){  
foreach (WaypointInfo wp in waypoints){  
if (wp.name.Equals(name))  
return wp.waypoint.position;  
}
return Vector3.zero;  
}
public bool IsAtDestionation(){  
Same function implemented  
in last lecture.  
.
..  
}
}
Exercise 2  
2
) Create a scene to represent the world specified in Exercise 1. Then,  
integrate the HSP Planner in the project and implement the actions  
of the NPC John to execute the generated plan.  
strong weapon (magic bow)  
weak weapon (sword)  
inside  
John  
weak troll  
Store  
chest key  
River  
chest  
strong dragon  
Castle  
Town  
Forest  
Cave  
Automated Planning in Games  
Games that are know for using planning algorithms:  
STRIPS-based action planning:  
HTN-based action planning:  
Automated Planning in Games  
There are many possible applications for automated planning  
in games:  
Planning NPC actions;  
Strategy planning;  
Design, test, and evaluate puzzles;  
Quest generation;  
Interactive storytelling;  
Hierarchical Generation of Dynamic  
and Nondeterministic Quests  
A combination of several story-related quests can be used to create  
complex narratives. The structure of the game's narrative can be  
represented as a hierarchy of quests.  
Lima, E.S. Feijó, B., and Furtado, A.L. Hierarchical Generation of Dynamic and  
Nondeterministic Quests in Games. International Conference on Advances in Computer  
Entertainment Technology (ACE 2014).  
Hierarchical Generation of Dynamic  
and Nondeterministic Quests  
Save family  
Take home  
Save wife  
PPrrootteecctt hhoouussee  
Escape  
Go to hospital  
Go to market  
Get antidote  
GGoo hhoommee  
GGiivvee aannttiiddoottee ttoo wwiiffee  
Kill wife  
¬
has(player,antidote)  
Ask old man for an antidote  
GGeett aannttiiddoottee  
Go home  
¬
has(player,antidote)  
Hierarchical Generation of Dynamic  
and Nondeterministic Quests  
Publications:  
Lima, E.S. Feijó, B., and Furtado, A.L. Hierarchical Generation of Dynamic and Nondeterministic Quests  
in Games. International Conference on Advances in Computer Entertainment Technology, 2014.  
Lima, E.S. Feijó, B., and Furtado, A.L. Player Behavior Modeling for Interactive Storytelling in Games. XV  
Brazilian Symposium on Computer Games and Digital Entertainment, 2016 [Best Paper Award].  
Lima, E.S. Feijó, B., and Furtado, A.L. Player Behavior and Personality Modeling for Interactive  
Storytelling in Games. Entertainment Computing, Volume 28, December 2018, p. 32-48, 2018.  
Further Reading  
Buckland, M. (2004). Programming Game AI by Example. Jones & Bartlett  
Learning. ISBN: 978-1-55622-078-4.  
Chapter 9: Goal-Driven Agent Behavior  
Millington, I., Funge, J. (2009). Artificial Intelligence for Games (2nd ed.).  
CRC Press. ISBN: 978-0123747310.  
Chapter 5.7: Goal-Oriented Behavior  
Further Reading  
Three States and a Plan: The A.I. of F.E.A.R:  
http://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear.pdf  
HTN Planning in Transformers: Fall of Cybertron:  
https://aiandgames.com/cybertron-intel/  
Planning in Games: An Overview and Lessons Learned:  
http://aigamedev.com/open/review/planning-in-games/  
Goal-Oriented Action Planning (GOAP):  
http://alumni.media.mit.edu/~jorkin/goap.html