AI2-THOR: A virtual environment for training home assistant robots

November 29, 2021

AI2-THOR is a framework based on Unity to train robots for a variety of Embodied AI tasks. Get a look into the design of AI2-THOR and how it is used to train robots that navigate, follow language instructions and understand the world around them.

Training robots is a computationally intensive task, and it can take robots days to learn simple tasks such as navigation. See how Unity Simulation Pro, Unity’s new simulation-optimized runtime build, has helped AI2 significantly reduce the training time.

This session first aired at the Unity AI Summit 2021 on November 18, 2021.

Presenter: Roozbeh Mottaghi, Research Manager, PRIOR team, AI2

Previous Video
Unity AI Summit 2021 keynote
Unity AI Summit 2021 keynote

Unity SVP of AI and Machine Learning Danny Lange shares how data-centric AI development can deliver success...

Next Video
Unity Simulation Pro: Scalable, cutting-edge simulations at your fingertips
Unity Simulation Pro: Scalable, cutting-edge simulations at your fingertips

Get the first look at Unity’s new simulation-optimized runtime build and how it can significantly boost per...