Sandhya Saisubramanian

Research Projects

AI for smart and connected communities

The goal of this project is to build intelligent decision support systems that reason under large, complex and stochastic real-world settings. We use a combination of sampling, optimization techniques, and game-theoretic approaches to achieve this. Our emergency response system and traffic patrolling systems have been deployed. News Article.

portrait portrait portrait portrait

Relevant papers: IAAI 2014, AAAI 2015, SPARK 2017, ITSC 2019.

Robust model reduction for risk-aware planning in large and stochastic environments

Many practical planning problems are intractable, even under full observability. We aim to design efficient model reduction techniques that balance the trade-off between model simplicity and risk awareness, to produce fast and high quality solutions. We employ a combination of heuristic search, sampling, and online planning to achieve near-optimal results, without compromising on the runtime.

portrait portrait

Relevant papers: AAMAS 2018, AISafety 2019, IROS 2019.

Reasoning under goal uncertainty

Many problems that require sequential decision making under uncertainty are goal-driven. However, it is often impossible to determine the exact goal states ahead of plan execution. The aim of this project is to design a framework that handles goal uncertainty, yet remaining tractable. Our framework introduces flexibility in goal specification by allowing a belief over possible goal configurations.

portrait portrait

Relevant papers: IROS 2019.