Open-World Police Pursuit Prototype

Project Overview

This project is a gameplay systems-focused open world prototype inspired by modern crime sandbox games. The primary focus of the project was designing scalable AI systems that could seamlessly transition between on-foot and vehicle-based pursuit while maintaining responsive gameplay behavior.

The project emphasizes:

  • AI pursuit systems

  • Vehicle chase behavior

  • Multi-state AI decision making

  • Vehicle possession transitions

  • Obstacle-aware driving logic

  • Gameplay system architecture

  • Debugging and system visualization tools

Police Pursuit AI System

  • The police pursuit system was designed to create dynamic chases where AI officers can pursue the player both on foot and in vehicles, depending on distance, context, and vehicle availability.

    The system supports:

    • On-foot pursuit

    • Vehicle pursuit transitions

    • AI vehicle entry and exit

    • Possession handoff between the officer and the vehicle

    • Obstacle-aware vehicle driving

    • Look-ahead steering behavior

    • State-driven pursuit logic

  • Player Detected

    ↓

    On-Foot Pursuit

    ↓

    Distance Threshold Reached

    ↓

    Locate Available Vehicle

    ↓

    Vehicle Entry

    ↓

    Controller Possession Transfer

    ↓

    Vehicle Pursuit

    ↓

    Obstacle Avoidance & Steering Updates

    ↓

    Return To On-Foot Pursuit When Necessary

Vehicle Pursuit & Steering System

  • The Vehicle Pursuit System uses predictive steering rather than directly targeting the player's current location.

    Instead of steering toward the player, the vehicle calculates a dynamic aim point based on the player's movement direction and steers toward that predicted location.

    This creates:

    • Smoother steering

    • Better corner handling

    • Reduced overcorrection

    • More believable pursuit behavior

  • The vehicle calculates an aim point ahead of the player's movement direction.

    Player Location

    +

    Player Movement Direction × LookAheadDistance

    =

    Aim Point

    The vehicle then steers toward the Aim Point instead of directly targeting the player.

  • To aid development and verify AI decision-making, custom debugging tools were implemented throughout the pursuit system.

    Debug Features:

    • Current AI State

    • Current Controlled Pawn

    • Distance To Player

    • Vehicle Entry Conditions

    • Vehicle Pursuit State

    • Dynamic Aim Point Visualization

    • Steering Direction Visualization

    • Obstacle Detection Debugging

  • These visualizations were used to verify that the AI was steering toward predicted movement locations rather than directly following the player.

    🔴 Red Line

    Displays the direct line between the police vehicle and the player's current location.

    🔴 Red Sphere Traces

    Three forward sphere traces project from the front of the vehicle to detect nearby obstacles and vehicles across the pursuit path. These trace results are used to determine avoidance direction and adjust steering while maintaining pursuit behavior.

    🟡 Yellow Sphere

    Displays the current aim point being used by the steering system.

Possession Transition System

  • A major technical challenge was maintaining AI behavior during possession transfers between the police officer and the vehicle.

    The system allows AI to:

    • Enter vehicles

    • Transfer controller ownership

    • Continue pursuit behavior

    • Preserve Blackboard state

    • Transition back to on-foot pursuit

  • Distance Threshold Reached

    ↓

    Move To Vehicle

    ↓

    Enter Vehicle

    ↓

    Controller Possesses Vehicle Pawn

    ↓

    Blackboard Updated

    ↓

    Vehicle Pursuit Branch Activated

Technical Challenges Solved

  • Possessing a new pawn can interrupt active AI behavior.

    To solve this:

    • Pursuit state information is updated before possession.

    • Blackboard values are synchronized before transferring control.

    • State transitions occur in staged steps rather than immediate possession swaps.

  • Early implementations could cause pursuit behavior to stop updating after entering a vehicle.

    This was solved by:

    • Separating vehicle and on-foot behavior states.

    • Explicitly controlling pursuit transitions.

    • Maintaining chase updates after possession transfer.

  • Directly steering toward the player created aggressive turns and unstable pursuit behavior.

    The solution was to:

    • Predict movement direction.

    • Generate dynamic aim points.

    • Continuously update steering targets.

  • Custom debugging tools were developed to monitor AI decision-making and verify pursuit behavior during development.

    These tools were used to validate:

    • State transitions

    • Vehicle entry conditions

    • Possession handoffs

    • Blackboard synchronization

    • Steering calculations

    • Aim point generation

    • Obstacle avoidance behavior

    The debugging system significantly improved iteration speed and allowed pursuit behavior to remain readable throughout development.