Master’s Thesis

Applying a Quantitative framework to affect pacing in FPS games

Under Construction

This is my master’s thesis where I built a Dynamic Difficulty Adjustment Model (DDA) for a simple 5 room dungeon in Fallout 4. The DDA will spawn items and enemies based on what the player interacts with and how well they are doing.

Goals:

Create a DDA that scales to both unskilled and skilled players
Make the player feel like every choice matters throughout the dungeon
See if I can map the traditional 3-act pacing curve 1:1 to a dungeon in Fallout 4.

Development Challenges & Solutions:

Abstracting threshold values into Global Variables

I initially had discrete integer values for increasing or decreasing difficulty. I realize this introduced a level of bias in that I as the developer would never be able to make a threshold value appropriate for all types of skill levels. (See Figure 1)
I decided to use a “Training Policy” where the system would intervene when the player reached 50% health or lower in any of the rooms.
This allowed me to create a binary system where the difficulty is either going up or down.
I then had an array of 10 bad guys. If the difficulty increased, the maximum number of enemies would be 6 bad guys and I would do an RNG calculation between 2-6 to spawn n number of bad guys.
This is still a threshold value in essence, but I was able to use global variables and abstract out how much the DDA was changing. (See Figure 2)

Figure 1: Unscabale Pseudocode

Figure 2: Pseudocode solution

Figure 3: Traditional 3 act pacing curve

Figure 4: Level Outline

What Went Well

I was able to achieve meaningfully different levels of pacing despite the varying skill levels of participants.

I am happy with the implementation of the training policy and the binary “difficulty up or down” solution.

What Went Wrong

I did not prove my hypothesis. The more I tried to make the DDA scale, the more hand tuning it required, which runs counter to what I set out to do.

This type of DDA implementation (hinging on performance) negatively impacts Explorers and Socializers. It also skewed my results.

I did not have a control group and told the player when the difficulty was going up or down. I think this warped some perceptions of the true difficulty.

What I Learned

I learned how to spawn items en masse via Enable Parent and/or Linked References. Despite requiring some hand placement, it was still a valuable workflow.

Always have a control group and consider the impact of the Observer Effect.