Lecture12 | Games Engineering

# Lecture 12 - AI
### SET09121 - Games Engineering

<br /><br />
Kevin Chalmers and Sam Serrels

School of Computing. Edinburgh Napier University

---

# Recommended Reading

- Artificial Intelligence for Games. Second Edition. Millington and Funge, 2009.
    - A good resource for anyone interested in game AI.

![image](assets/images/physics_book.jpg)

---

# History of Game AI

What is Artificial Intelligence?
- Artificial Intelligence (AI) is the study of the intelligence of machines, and the attempt to replicate human-like intelligence in a machine.
    - This is a very wide area of study, incorporating not only technical implementations of intelligence, but also ethics and philosophy.
- From a games point of view, we will take a very simple outlook:
    - AI is any algorithm or code that controls the behaviour of one of our game entities.
- Therefore, our Pong example back at the start of the module had a form of AI.
- It was dumb, but it still behaved in a way that would allow it to be competitive.

---

# Academic AI vs. Game AI

- Academic AI can be (very roughly) broken down into two phases:
    - Symbolic AI (early days)
        - Set of knowledge and reasoning algorithms.
    - Nature inspired and learning (modern era)
        - Inspired by models in nature and statistical inference.
        - Used in learning, neural networks, genetic algorithms, etc.
- Game AI is still rooted in the deterministic, classical, symbolic AI era.
    - Symbolic AI still used extensively in games.
        - Path finding, state machines, etc.
    - Modern techniques have been tried, but are rarely successful.
    - Modern techniques have seen more success in content creation.

---

# In the beginning...PacMan

- PacMan is recognised as one of the earliest examples of AI in games.
- PacMan AI relied on state machine behaviour.
    - We will look at state machines in a separate lecture.
- This was in 1979...
- ...game AI did not change much until the mid 1990s.

![image](assets/images/pacman.gif)

---

# Timeline

- In the mid-1990s AI started to become a selling point.
 - **1994**: Beneath a Steel Sky mentions AI on the box.
 - **1997**: GoldenEye 007 introduces world sensing, allowing enemies to see allies and notice when they were killed.
 - **1998**: Thief: The Dark Project and Metal Gear Solid expanded the world sensing concept.
- Real-time strategy games also emerged in the 1990s.
 - **1994**: Warcraft used noticeable path finding techniques.
 - **1998**: Warhammer: Dark Omen has robust formation motion.
- Games also start emerging where AI is the main game mechanic.
 - **1997**: Creatures.
 - **2000**: The Sims.
 - **2001**: Black and White.

---

# AI Techniques for Games

- There are numerous usable AI techniques applicable for games.
    - Classical AI techniques -- common.
    - Newer academic techniques -- uncommon.
- Different techniques accomplish different aspects of behaviour.
    - Movement of entities.
    - Decision making for entities.
    - Strategic (planning) decisions.
    - Learning from player behaviour.
- We will only look at the first two in detail as they are the most common. Other modules look at learning and planning in a general AI context.

---

## Movement

---

# Movement and Steering Behaviours

- Steering behaviours.
    - An algorithm which determines how an entity should move.
    - Can be goal based:
        - Seek, flee.
    - Or can be more general:
        - Patrol, wander.
- Steering behaviours are an the base of most game AI.
    - It allows us to move entities in a certain manner.
- Steering behaviours can also be combined to create more complex behaviour.
    - Look into flocking behaviour to get an idea.
- We will look at steering behaviours next week.

---

# Path Finding

- Path finding is the discovery of the route between two points in a game world.
- Information from path finding can be fed into steering behaviours.
- Two core techniques:
    Waypoints: :   most common.
    Navigation mesh: :   becoming more popular (see image).
- Two common algorithms:
    - Dijkstra (slow but checks everything).
    - A* (fast).

![image](assets/images/waypoint-mesh.jpg)

---

## Decision Making

---

# Decision Making

- We are going to look at two types.
- State machines.
    - Similar to the idea of state modelling in UML.
    - Character has a number of possible states.
        - e.g. attack, hide, run, etc.
    - Character determines current state based on any number of conditions.
    - Character will change state when a particular action occurs.
        - If patrolling and player is spotted then change state to attacking.
- Decision trees.
    - Similar to activity diagrams in UML.
    - Used to control characters decision making process.
    - Can also be used to control animation.
    - Very simply AI technique to implement but can be very powerful.

---

# Decision Making

- Fuzzy logic
    - Logic design with grey (fuzzy) areas.
    - Fuzzy logic applies some randomness to decisions.
    - It tries to be more in line with how people think.
    - Not very popular in academia any more -- probability and statistics preferred.

![image](assets/images/fuzzy-logic.png)

---

# Decision Making

- Some others...
    - Behavioural trees
        - Chaining tasks and decisions together to form complicated behaviours.
    - Markov systems
        - Similar to fuzzy logic but works with probability or priority of transitions.
    - Goal-oriented behaviour
        - Character chooses an action based on its current goals.
    - Rule-based systems
        - Database of "if" conditions to determine the behaviour to take.

---

# Strategy - World Data

- Modern game AI techniques rely on data from the game world to be effective.
    - We have already seen this with path finding.
- Depending on the AI technique used, different types of data can be required.
    - Movement may need to know about obstacles, jump points.
    - Knowledge of cover.
    - Knowledge of other characters in the game world to launch attacks.
    - etc.
- Most of the techniques requiring world data are referred to as strategic techniques.

---

# Strategy - Waypoint tactics.

- Areas of the map are marked for tactical significance.
    - For example cover positions, sniper positions, etc.
- The AI determines which waypoint to head for based on an algorithm.
- This allows strategic looking behaviour from the game characters.
- This technique is used extensively in FPS and similar games.
- Tactical path finding.
    - Takes consideration of the surroundings when determining how to move between waypoints.

---

# Strategy - Waypoint tactics.

---

# Tactical Analysis

- Essentially a more complicated version of the previous ideas.
- The map is marked up with values to help determine where characters should move to.
- An algorithm analyses the date to determine the action to take.
- For example:
    - Consider a RTS game.
    - There is a base located on the map.
    - The base area is given a high target value.
    - This makes the game AI try to control this base area.
- Adding dynamic data means that the AI will dynamically respond to the player's actions.

---

## Learning

---

# Learning

- Neural networks are a popular nature inspired technique. 
- They are modelled on a simplified idea of the brain.
    - Neurons input a signal.
    - Output signal to other neurons.
- Can have learning developed over time.
- Academia and hobbyists see it as a method of believable learning.
- So far fallen short on game projects (see Creatures, Fable II).

![image](assets/images/neural-network.png)

---

# Learning

- Emergent behaviour.
    - Evolutionary algorithms.
        - AI is given a set of values (genes) to determine the likelihood of performing an action.
        - If the AI is successful the genes are used for breeding the next iteration.
    - Bio-inspired algorithms.
        - Ant colony optimisation.
        - Artificial immune systems.
- Depending on your programme you might study some of these topics:
    - Multi-agent systems.
    - Computational intelligence.
    - Emergent computing for optimisation.

---

# Goal of AI -- Realism

- The goal of game AI is to provide a challenge to the player and create a realistic, living world.
- Examples:
    - GTA IV
    - Assassin's Creed Origins
- Notice:
    - People walking around.
    - Reaction to environment effects (e.g. rain).
    - Enemies attacking in realistic looking manners.
    - etc.

---

# Goal of AI -- Realism

---

# Goal of AI -- Realism

---

## Bad AI

---

# Bad AI

---

# Bad AI
<iframe width="1400" height="800" src="https://www.youtube.com/embed/bc5BwK3iNh0" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen=""></iframe>

---

# Bad AI - Path finding problems
<iframe width="1400" height="800" src="https://www.youtube.com/embed/lw9G-8gL5o0" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen=""></iframe>

---

# Bad AI

- When the game designer comes up with an idea, they have a vision of how the game will play.
- A game is meant to be a carefully crafted experience for the player.
- AI can and does bring unpredictability to the game.
    - The game designer might not want unpredictable behaviour.
- Developers may override the AI decisions for the sake of gameplay.

---

# Bad AI

- There are lots of examples of bad AI breaking immersion.
- Stupid NPCs getting stuck in corners or getting in your way.
- The design of the level needs to take account of the NPCs' ability to navigate it.
    - Hence you normally get large doorways and corridors.
- Unfortunately, the AI, NPCs, and levels are normally designed by different teams.
    - So it doesn't normally work first time.
    - At some point comprises have to be made.
- If the NPC is on the player's side you need to make sure it does not let them down.
    - Halo was notorious for this. Fallout 4 has problems also.

---

# Cheating AI

- Many game AI systems cheat.
    - Pro Evolution Soccer
- Sometimes this is referred to as rubber band AI.
    - A term originally coined in racing games.
    - Opposition always appeared to keep up with you no matter how well you drove.
    - Opposition was considered to be on a rubber band attached to the player.

---

# Cheating AI

---

# Summary

- We have looked at a very broad picture of what game AI is.
    - Essentially, use classical, deterministic techniques.
- We also introduced some techniques that are used in games.
    - Movement
    - Decision making
    - Strategy
    - Learning

- And how AI can sometimes not work how the player wants.
    - Bad AI
    - Cheating AI
- All these areas are worth considering depending on your game. Just think about the time you have and the experience you are aiming for.