How We Used Reinforcement Learning to Solve the Abbey of Crime

Overview

Explore how AI and reinforcement learning were used to conquer the challenging 1987 RPG "Abbey of Crime," overcoming its complexity within 120k of data to complete the game autonomously.

Syllabus

• Introduction to the Abbey of Crime
Overview of the 1987 RPG and its complexity
Game challenges and objectives that need addressing with AI
• Fundamentals of Reinforcement Learning (RL)
Key concepts: agents, environments, states, actions, and rewards
Overview of RL algorithms used in gaming
• Analyzing the Abbey of Crime for RL Application
Mapping game elements to RL structures
Challenges in modeling game dynamics with RL
• Designing the RL Framework
Setting up the environment and state representation for Abbey of Crime
Action space definition based on game mechanics
• Implementing RL Algorithms
Selection of appropriate RL algorithms for the game
Training and tuning parameters to achieve desired outcomes
• Handling Game Constraints and Data Limitations
Strategies for efficient data use within 120k constraints
Optimization techniques for resource-constrained RL applications
• Testing and Evaluation
Methods to evaluate RL performance in-game
Iterative testing and refinement processes
• Achievements and Outcomes
Discussion on solving complex game scenarios
Insights and breakthroughs from conquering the Abbey of Crime with RL
• Lessons Learned and Future Directions
Reflections on the role of AI in gaming
Potential applications of RL across other gaming contexts
• Conclusion and Resources
Summary of key learnings from the course
Further reading and resources for advanced exploration of RL in games

Subjects

Conference Talks