Frits Lyneborg COPENHAGEN

Unreal developer with a bacground in Unity and Blender

Developing games for various studios since 2014

I work as Technical Artist , Technical Animator as well as Full Stack Game Developer

Mail me LinkedIn
 FSM based AI

Interactive AI instead of hardcoded animations

When building your AI from ground up with physics and Finite State Machines, you can get incredibly live and all-situations-handling AI.

Players often think it's Animated: "Just like that Pixar Logo".

It could not be further from the truth.

This is live, this is not animations. These are characters you can play with in a game. The Pixar-lamp is just an animation, it can not react with anything a player does.

Stress Testing

100% physics driven animations can take longer time to develop than ordinary animations, but the time spend is usually rewarded many times in the later development stages

  • When done right, assets will always perform well, no matter what stress they find themselves in
  • If playtests shows issues in edge-cases, it is usually a creative challenge to overcome, instead of some ugly animation-hack
  • They are easy to stress test: Just spawn a lot into a chaotic scene: They should always perform well, no matter what occurs
 Programming the Physics Engine

Drifting AI is fun AI

Assets you can throw into any chaotic stressed game scene gives game designers creative freedom.

Opposed to many programmers who prefers "to be in control as much as possible", I just absolutely love to make drifting, uncontrolled stuff!

It doesn't even have to be a complex programming task when making good use of the Physics Engine.

When speed in development is of essence, and all you need is something that looks like if someone is driving the other cars, a few lines of code can get us as far as these ball playing drifting AI cars:

Playing ball with the player

  • Note how the cars drive up to the ball, wait for the right moment, then speed up to kick it.
  • The cars are controlled by one simple script, no Machine Learning
  • Any type of chaos (like the green boxes, hills and slopes) is handled
  • The cars are aware of other nearby cars via triggers, and can perform any flock behavior, aggressive takeovers or what is needed for the game
Intelligent use of AI

Machine Learning drivers?

Machine learning is unfortunately completely useless for any type of NPC - despite the hype

Neural Networked Machine Learning Agents are able to do anything from driving cars to Speech recognition, so many people think this form of AI is also the next big thing when making games.

As an example here's Unity's praised tutorials, in which they teach people how to use Machine Learning Agents to drive NPC's in a racing game.

When developing NPC's, AI, interactive Game assets, you want:

  • Short times pan from idea to test
  • Simple projects
  • AI you can mold to your needs during an iterative development phase

- Basically everything Machine Learning is not!

Machine Learning is (surprisingly) useless in actual real life game development!

The situation is similar to when people figured that 3D printers would end world hunger..

(Just like some people said "Smart new technology, now you can just print the food" - some people think "Smart new technology, now we can just train agents to be NPC's") 

Mobile Friendly | Game friendly

Rule based AI makes the best gameplay

Swarm bots do not have to be only Chaotic or Predictable.

In many situations it's most fun for the players of a game, if AI appears chaotic at first, but then they realize that they can "trick" the AI, because it is somewhat predictable.

This should IMHO be accounted for and encouraged, to support emergent gameplay.

As a bonus this mindset often also leads to more readable, controllable code for us as developers. Have a look at this example. These are not the overly complex "Boids" algorithms.

The way these navigate is actually just a simple hack: Not visible for the player is a zero-gravity ball flying around in some obstacles. They just either follow that, or the visible green circle.

 Animations and physics mixed

Mixing regular animations with Physics AI

When you want something to animate, but still be highly interactible

The running robots in this video are animated, but at the same time they are wrapped in physics, so they actually react to any situation, for examle when having beach balls smacked into their face.

If this was pure enimation-based, the reactions would monotome. If it was pure Physics driven AI, it would lose a game-feel.

 Stomping on a beat

Precision in Chaos? Not a problem!

You can design things to happen predefined in shifted time, even in a chaotic, interactive environment

In this little demo, I wanted to show how a 100% physics driven world does not have to be sloppy and out of control, even though everything is fully interactible and emergent.

(Please enable sound for this demo)

Note how it is not the Jumping up in air that is on beat, but actually the later landing.

The rabbits performs this little trick sharply on beat, landing in pose no matter if they are thrown off course, or when they are spawned:

Everything will self-correct.

Tricks like this gives a Game Designer a ton of freedom for designing interesting gameplay!

Real life project?

Well, I did do this in 2008 when I was into building robots ;)

 Background in hobby Robotics

My background in Hobby Robotics

Back in 2008 I founded the worlds largest community for hobby robot builders

Together with millions of Very Talented Persons from all over the world, we made thousands of fun robot- and AI projects

Here's a drifting car video I did for Make Magazine in 2011 ;)