Explore the journey of creating GT Sophy, a superhuman racing AI agent, in this 24-minute conference talk by Alisa Devlic, Senior Research Scientist at Sony AI. Discover how deep reinforcement learning techniques were used to develop an AI capable of mastering race car control, tactics, and etiquette to compete against the world's best Gran Turismo drivers. Learn about the technical evolution of GT Sophy from a research project to an in-game feature in Gran Turismo 7, including novel training and evaluation methods implemented on a modern cloud computing platform. Gain insights into the collaboration between Sony AI, Polyphony Digital, and Sony Interactive Entertainment that led to this breakthrough AI agent becoming a permanent feature in GT7 less than two years after its Nature journal cover appearance.
From Hypothesis to Reality: Designing a Superhuman Racing AI Agent Using Deep Reinforcement Learning
Overview
Syllabus
From Hypothesis to Reality: Designing a Superhuman Racing AI Agent Using a Deep RL by Alisa Devlic
Taught by
GAIA
Related Courses
-
AI skills: Introduction to Unsupervised, Deep and Reinforcement Learning
-
Outracing Champion Gran Turismo Drivers With Deep Reinforcement Learning
-
WarpDrive: Orders of Magnitude Faster Multi-Agent Deep Reinforcement Learning on a GPU
-
Deep Reinforcement Learning for Atari Games Python Tutorial - AI Plays Space Invaders
-
Introduction to Reinforcement Learning
-
Building Adaptive Multi-Agent AI Systems: From Theory to Implementation
