Simulated environments have been essential to advancing the field of artificial intelligence, providing vast amounts of synthetic data that tests novel approaches safely and efficiently. This has most often taken the form of games, ranging from simple board games to modern multiplayer strategy games.

These games served as a good starting point, but Danny Lange (Unity Technologies) reveals an opportunity to push the state of the art in AI research to the next level. United introduced the Obstacle Tower, a high-visual-fidelity, 3-D, third-person, procedurally generated game environment purpose built to test a deep reinforcement learning-trained agent’s vision, control, planning, and generalization abilities. Over the past year, Unity invited researchers and developers to try to solve the tower with the intention of sharing those insights with the broader community.

Prerequisite knowledge

• A basic knowledge of machine learning and AI

What you'll learn

• See how you can use what Unity learned from hosting the challenges to engage the broader community to advance AI research
• Find out how participants fared as they attempted to solve the tower, what that taught Unity, and what's next for Obstacle Tower as it continues pushing advances in deep reinforcement learning
• Learn how observing emergent behaviors of multiple AI agents in a simulated virtual environment can lead to the most optimal designs and real-world practices

This session is from the 2019 O'Reilly Artificial Intelligence Conference in San Jose, CA.