The AI's patrol behavior has two different modes:

1. Random Patrol: This mode operates autonomously without level designer input. The AI selects random destinations within a specified range that are confirmed to be on the navigation mesh, then moves to those locations. This provides a simple patrol pattern that requires no manual setup.

2. Defined Pathway Patrol: This mode follows a specific route designed by the level designer. The level designer creates a sequence of locations that forms a patrol path. While this offers more precise control over AI movement patterns, designers must be cautious when placing these waypoints to ensure they're accessible and create logical paths.

The key aspects are:

• The AI perception relies mainly on hearing stimulus rather than sight stimulus for detection, as it has a limited visual range. 

• Whenever a new sound occurs in its hearing range, it will move to the sound location to investigate. If there is nothing to detect at that location, it will get confused and go back to the Patrolling State.

• If the player is running, their footsteps will continue generating noise, causing the AI to follow/chase

• The player can break this chase pattern by switching to walking or crouching, which generates less or no noise. Or the player can throw a brick to distract the AI.

• If it sees the player, it will go to "Aggressive State"

This design creates an interesting stealth dynamic where players must manage their movement style - running draws constant attention while sneaking allows them to evade detection.

The AI uses Unreal Engine's built-in AI Perception system to process both visual and auditory inputs from the environment. Here's how it works:

The perception update function handles the core logic:

• Every time the AI detects something (sight or sound), it stores that detection information

• For sound detections:

  • The AI compares the volume of the new sound against previously recorded sounds

  • If the new sound is louder, it becomes the AI's new target point to investigate

• For visual detections:

  • As soon as the AI visually spots the player, it immediately switches to an aggressive state

  • The behavior changes from "Investigation State" to "Aggressive State"

The AI executes these actions every frame within its update loop:

• Emits a scream to indicate it has engaged with the player for the first time in the current state

• Determines the player's current location in the game world

• Navigates toward that player's position

• If within attack range, perform an attack on the player

However, we must handle edge cases where the player moves outside the navigation mesh. In such situations, the AI could get stuck in its aggressive state while trying to reach an unreachable position. To prevent this, we've implemented a time limit that forces the AI to exit the aggressive state after a certain duration. 

The attack sequence consists of three stages:

1. Lifting: The AI grabs and lifts the player, disabling all player input and forcing them to face the AI

2. Holding: Player controls are restored, allowing them to attempt to escape through rapid key presses

3. Throwing or Killing: Based on the player's escape attempt

   • Successful Escape: The AI throws the player forward and becomes stunned temporarily, giving the player time to escape. The AI then loses track of the player and returns to patrol

   • Failed Escape: Results in player death

The entire sequence is implemented using Unreal Engine's Animation Montage and Animation Notify States, which triggers code execution at specific keyframes in the animation. This synchronization between code and animation creates fluid, well-timed interactions that enhance the gameplay experience.

I created both the opening and closing cinematics for the game using Unreal Engine's Sequencer tool. This allowed me to capture choreographed character movements within the game environment and manipulate shader parameters to create camera-based visual effects. I collaborated with the level design team to ensure these sequences aligned properly with the audio.