Ishan Durugkar

Currently I am a Research Scientist at Sony AI working with the Game AI team to bring reinforcement learning agents to games such as Gran Turismo.

I completed my PhD at the Learning Agents Research Group LARG advised by Prof. Peter Stone. My research focuses on the sub-field of Machine Learning called Reinforcement Learning (RL). My doctoral thesis focused on improving RL algorithms using the estimation and control of the visitation distributions induced by reinforcement learning agents, such as the distribution of states they visit, or the distribution of transitions they experience.

I have interned at DeepMind (Summer 2021) with Volodymyr Mnih, and Microsoft Research (Summer 2018) with Matthew Hausknecht and Adith Swaminathan.

Apart from the above highlights, I have worked on a variety of projects that looks at different aspects of the RL problem. Some of these directions that I have worked on are optimization perspectives of TD learning, considering agent preferences in multi-agent RL, using adversarial techniques for transfering policies from simulation to real robots, and some applications of RL (for example, learning a policy to navigate knowledge bases). Take a look at my Publications page for an up to date list of my papers.

I also coordinate the weekly RL Reading Group. Contact me in case you are interested in joining in our discussions.

Finally, I also participate in the RoboCup Standard Platform League as part of the UT Austin Villa team. Here I focus on working with real robots. Apart from trying to get research like sim-to-real transfer applicable practically in this competition, this experience has helped me really understand robotics systems hands on, along with multi-robot communication and coordination.

Previously, I have been a Master’s student at UMass Amherst and have worked with Sridhar Mahadevan at the Autonomous Learning Lab on both Deep Learning and Deep RL projects.