Prof. Doshi's research interests broadly fall in AI and Robotics. In the area of AI, he is an expert on autonomous decision making with specific interests in decision making under uncertainty in multiagent settings. In robotics, Prof. Doshi investigates ways to make learning by observing pragmatic for robots and is an expert on inverse reinforcement learning. He also studies methods for SLAM in occluded, multi-robot settings. His past research experience also includes the semantic Web and specifically in ontology alignment and learning; and in services-oriented computing, specifically in composing Web services and adapting the compositions.
In collaboration with Prof. Piotr Gmytrasiewicz at UIC, Prof. Doshi co-pioneered the Interactive POMDP (I-POMDP) framework, which complements the predominant focus of previous multiagent research on team decision making. I-POMDP departs from several traditional game-theoretic solution concepts (such as equilibria) and its subjective perspective permits a natural consideration of issues related to interactive epistemology (nested modeling) and computability (finite nesting) in decision making. I-POMDPs are now well recognized within the multiagent community as a leading framework for decision making in complex, general settings. Recent use cases of I-POMDPs by researchers testify to its significance and growing appeal. They are being used to explore strategies for countering money laundering by terrorists, enhanced to include trust levels for facilitating defense simulations, and building empirical models for simulating human behavior pertaining to strategic thought and action. Survey articles published by Prof. Doshi in the AI Magazine and the AI Journal offer easy readings for a contextual understanding of this framework. THINC Lab also maintains a one-stop repository of all papers related to the I-POMDP framework. In 2011, Prof. Doshi received UGA's Creative Research Medal for his work related to I-POMDPs, which acknowledges exceptional achievements in creativity and research by UGA faculty.
Prof. Doshi would like to see robots learn tasks simply by observing others perform them. Toward this ambitious goal, his research investigations focus on generalizing inverse reinforcement learning (IRL) to operate in contexts involving noisy sensor models and where portions of the observed task may be occluded from view. A recent survey article published by him and his doctoral student offers an informative review and comparison of various IRL methods and their extensions. This research is being evaluated by teaching collaborative robotic manipulators on a produce processing line to accurately sort onions.
Prof. Doshi is the recipient of the 2009 NSF CAREER award for his research on multiagent decision making. His sustained research excellence has earned him the Outstanding Faculty Research award from the CS department three times (2009, 2012, and 2018). He has published extensively in journals, conferences, and other forums in the fields of agents, AI, Robotics, Semantic Web, and Web services with over 150 archival publications. He has given numerous presentations in conferences and invited talks at research institutions and universities. His papers are available from this website's publication page or from his Google Scholar profile. He currently serves on the editorial board of Springer's Journal of AAMAS as a coordinating editor and as the area chair in various AI conferences.
M.S. (Computer Science), Drexel University, 2001
Ph.D. (Computer Science), University of Illinois at Chicago, 2005.
Artificial intelligence & Robotics
Decision-Making under Uncertainty, Multi-Agent Systems
Reinforcement Learning, Learning from demonstrations