Week 1- Research and Exploration

Group Behavior is often overlooked, but it is there in almost every game.

Horizon Zero Dawn’s Hierarchical Network Planner is some of the most impressive Game AI work I have come across.

Hordes of Zombies that attack you in L4D and other PvE games- they all have an essence of group behavior which is the key to the combat sequences.

But I wanted to shed light on a different kind of group behavior- in city builders.

Airport Tycoon, Rollercoaster Tycoon and Prison Tycoon were how I spent my time in the hot, humid Mumbai summers. Being god and just popping up a roller coaster with a click was all the fun my juvenile 12 year old brain needed.

RCts-l300.jpg

These games evolved a lot. Lets look at the 1999 Roller Coaster Tycoon Game for example: https://photos.app.goo.gl/tL8gbfd6d191HGyFA

The agents move randomly and their behavior is almost ant-like. One glaring that caught my eye is that people visited my theme park ALONE ):

This caught my eye to research group AI algorithms.

Later, I saw some Planet Zoo footage, and this 2019 game is a clear evolution of the Classic Roller Coaster Tycoon. One notices flock behavior with steering and people visited the zoo in groups. This is what brought the park to life.

These tycoon games are grounded in reality. You do things meant for humans and the “juice” is purely dedicated towards a human and the AI should play a crucial role. Somewhere in this dev pipeline, the RCT from 1992 with scatter people roaming around like scattered bugs evolved to something like Planet Zoo where people actually walked in groups.

I saw myself asking questions like

  • How do agents know what “group” they are assigned to?
  • How do agents know to comeback to the group if they are separated?
  • How do group based navigation systems work?
  • And so on…

Game AI Pro is a great read to understand AI systems further and I spent a whole week getting into some of these chapters.

The two readings that caught my eye were:

1000 NPCs at 60 FPS – Robert Zubek  Chapter 34

  • Talked about how actions should be “routines” to better understand AI agents
  • How to optimize AI agents by delegating them a set of tasks

Hierarchical Architecture for Group Navigation Behaviors – Clodéric Mars and Jérémy Chanut – Chapter 20

  • Talked about flock behavior
  • Decision Making in Flocks

The second reading is what I will be using as a major influence for establishing goals for my further endeavors.

Planned Goals:

  • Multiple groups patrolling on a map
    • Every group has 4 agents
    • Map has a “food store”
  • Every agent has a “hunger” stat which deteriorates every second
  • When hunger stat for an agent is low they break the formation and go to eat food but the group has to continue in formation
  • After hunger stat is full, agent goes back to group and continues to be in formation
  • Stretch Goal: Introduce another stat and make the agents think in formation
  • Stretch Goal#2: Sub groups within groups

Stay posted on my second week of development as I hit some of those goals and come across obstacles. Feel free to contact me with any more ideas!

 

%d bloggers like this: