Duration: 8:01

1,459 views

10

Developing and testing real-world AI is an expensive and time-consuming process. We need to collect a large amount of annotated training data in a variety of conditions and environments, and such data-driven systems can result in failure cases that can jeopardize safety. In this session, we will explore how high-fidelity simulations can help us alleviate some of these problems. We will discuss how near-realistic simulations can help not only with gathering training data, but also can be embedded in imitation-learning or reinforcement learning loops to improve sample complexity. Our discussion will center around AirSim, an open-source simulator built on Unreal Engine that offers physically and visually realistic simulations.

See more at https://www.microsoft.com/en-us/research/video/high-fidelity-simulations-the-critical-path-to-real-world-ai/