A Competitive Time-Trial AI for Need for Speed: Most Wanted Using Deep Reinforcement Learning

Explore a 44-minute conference talk from the 38th Chaos Communication Congress (38C3) detailing the development of a competitive AI system for Need for Speed: Most Wanted (2005) using deep reinforcement learning. Learn how a former eSports player transformed his passion for racing games into an AI development project, covering the complete journey from game hacking to create a custom API, building an OpenAI gym environment, implementing virtual controller steering, and successfully training an AI using the Soft-Actor-Critic algorithm. Discover the technical challenges and solutions involved in creating a Python-based open-source AI system capable of competing in time trials, demonstrating how modern hardware capabilities and software frameworks made this long-awaited project possible.