Raman Arora, Johns Hopkins Whiting School of Engineering
Ryan Gardner, Johns Hopkins Applied Physics Laboratory

Deep reinforcement learning (DRL) is an emerging family of machine-learning techniques that enables systems to learn complex behaviors through interaction with an environment. This project’s goal is to design online learning agents that avoid undesirable outcomes, promising broad impact on real-world applications ranging from national security to civilian systems.

The project creates a novel learning framework that integrates sensitivity to risk and applies deep reinforcement learning to real-world autonomous systems, taking a game-theory view of the learning problem. Developing the framework will present two main challenges: designing a learning agent with an appropriate, risk-sensitive reward function and understanding the limitations of online learning and sequential decision-making in a dynamic, possibly adversarial environment.

Ultimately, this work will create a unique DRL capability that can reduce risks and increase reliability of real-world autonomous systems, with broad applications to multiple domains including healthcare, autonomous vehicles, and smart cities.