Deep RL-enabled Fortnite Agent

Overview

This ongoing team project, conducted under the Data Science Group, focuses on developing an intelligent agent capable of playing the highly complex battle royale game “Fortnite” using state-of-the-art deep reinforcement learning techniques.

Project Details

Duration: October 2024 – May 2025
Affiliation: Data Science Group, IIT Roorkee
Role: Team Member

Technical Approach

Our approach combines several cutting-edge techniques to tackle the multi-faceted challenges of Fortnite gameplay:

Core Technologies

• Multi-linear ICVF (Implicit Conditional Value Functions): Implementing temporal difference learning for value-based decision making
• DreamerV3: Utilizing as the world model to predict future states and plan optimal actions
• NVIDIA COSMOS: Employed as the embedding model to process and embed video streams for training

Challenges Addressed

1. Complex Action Spaces: Managing the vast array of possible actions in Fortnite’s dynamic environment
2. Multi-agent Dynamics: Handling interactions with multiple opponents in real-time
3. Long-term Planning: Balancing immediate rewards with strategic long-term objectives
4. Visual Understanding: Processing complex visual information from game streams

Technical Implementation

The project integrates:

• Advanced computer vision techniques for game state understanding
• Sophisticated reinforcement learning algorithms for decision making
• Real-time processing capabilities for responsive gameplay
• Multi-modal learning approaches combining visual and strategic information

Innovation

This project pushes the boundaries of RL applications in complex, dynamic environments and contributes to the advancement of AI agents in competitive gaming scenarios.

This project represents a significant challenge in applying modern deep reinforcement learning to one of the most complex and dynamic gaming environments available today.