Research Style of Sven Koenig

I am an Artificial Intelligence (AI) researcher who believes that situated agents (such as robots or decision-support systems) must be able to make good decisions in complex situations that involve a substantial degree of uncertainty, yet find solutions in a timely manner despite a large number of potential contingencies. My research is intended to help provide a strong foundation for building such agents.

We currently live in an era of "big data," since our online and other activities result in lots of collected data, and machine learning is used to obtain models that help us to make sense of the data. Ultimately, one wants to use these models to make good decisions. Most of my research therefore focuses on decision-making methods that enable single agents and teams of agents to act intelligently in their environments and exhibit goal-directed behavior in real-time, even if they have only incomplete knowledge of their environments, imperfect abilities to manipulate them, limited or noisy perception or insufficient reasoning speed. I develop new decision-making methods and study their properties theoretically and experimentally, often using problems from robotics (or video games) and/or in collaboration with robotics researchers. I also study decision-making methods developed by other researchers to determine how good they are and when they should be used. While my research results may appear diverse, there is a common underlying thrust, namely to bring about advances that extend the reach of search (in a broad sense), including heuristic search, planning and optimization in general. Such methods apply to almost all decision-making problems ("everything is search"). Yet, there are different communities that study search, for example, in AI, robotics, theoretical computer science, decision theory and operations research, often in isolation. I am trying to bridge this gap both by making search methods more powerful and thus more useful (pushing emerging classes of search methods) and by applying them to novel domains. I am a very collaborative researcher, with current research collaborations not only in the US but also in Australia, Canada, Chile, Germany and Israel.

Not surprisingly, I believe that the key to making progress in my research area is to combine ideas from different decision-making disciplines, which requires serious technical advances to reconcile the different assumptions and approaches in a way that results in synergy between them. I pursued degrees in business administration (taught with a strong focus on applied operations research) and AI precisely because I was interested in the decision-making methods that different disciplines have to offer. This background also enables me to build up collaborations with researchers outside of AI. As described below in more detail, my collaborators and I combined ideas from different decision-making disciplines, resulting in novel insights and methods. For example, we combined ideas from AI planning and utility theory to create the algorithmic foundations for building decision-support systems that fit the risk preferences of human decision makers in high-stake decision situations better than existing systems; we combined ideas from robotics and theoretical computer science to analyze common robot-navigation methods in known and unknown terrain (such as goal-directed navigation, coverage, localization, mapping, pursuit evasion and multi-agent path finding), resulting in a better understanding of their quality; we combined ideas from heuristic search and algorithm theory to develop re-planning methods in highly dynamic domains that, different from the ones that had typically been studied in AI, are able to provide guarantees on the resulting solution quality; we combined ideas from robotics and economics to build teams of agents that use cooperative auctions to distribute tasks autonomously among themselves; we combined ideas from real-time heuristic search and robotics to build terrain-covering ant robots that promise to result in fault-tolerant navigation behavior for teams of minimalistic and thus cheap robots; and we are currently combining ideas from AI, optimization and robotics to create the algorithmic foundations for the next-generation of automated warehouses to reduce the delivery times of ordered goods.

Accordingly, I try to publish broadly across different parts of AI, such as in conferences on AI in general (IJCAI, AAAI), planning (ICAPS), autonomous agents and multi-agent systems (AAMAS) but also constraint programming (CP), reasoning with uncertainty (UAI) and knowledge representation and reasoning (KR), and more narrowly in robotics (ICRA, IROS but also RSS).

International Conference on Automated Planning and Scheduling 2006

Home Page of Sven Koenig