IST Lunch Bunch
Building AI powered flying machines with high fidelity simulations
Abstract: Machine Intelligence techniques such as reinforcement learning and imitation learning have been tremendously successful in achieving superhuman performance in software games (e.g. Atari games, AlphaGo etc.). While the idea of applying such methods for autonomous flights in an uncertain environment is an appealing one, the high sample complexity of these techniques adds a unique challenge for the domain. In this talk, we will discuss different ways to address this challenge. First, we propose novel algorithms that exploit the statistical structure of uncertainty in order to rapidly build accurate prediction engines. Second, we also explore the feasibility of building and using high-fidelity simulation environments that mimic the real-world. The key idea being that such a simulated world would provide the necessary experience and generate enough data to train the system adequately before deploying it on a real aircraft.
Contact: Diane Goodfellow at 6267972398 firstname.lastname@example.org