O3DCon

Simulation software is becoming increasingly more central to building powerful new AI systems that offer to change the world, and game engines are uniquely positioned to become a foundation for these simulations. However, open-source simulation software is needed to help democratise powerful AI and make it available to a diverse community. Conor McMullan, Director at invidy AI, demonstrates the untapped potential O3DE has to train useful AI in multiple complex scenarios. Taking advantage of O3DE’s Gem API and shader technology for real-time inference, participants get to feel like they’re back in middle school by playing against an adept AI-powered dodgeball team. Next, Conor demonstrates how to train drone AI within O3DE’s engine, specifically in fighting wildfires. O3DE’s built-in Gem and shader technology enable in-engine network training- trends and results can be immediately visualised, debugged and understood. Finally, Conor offers a technical breakdown of their approach and their philosophy of building on the engine's strengths. Following the session, invidy AI will release the AI training environment under the MIT license for the O3DE community.