Scalable planning under uncertainty with repeated graph searches


These projects aim at developing algorithms that can solve large-scale planning under uncertainty problems by utilizing graph searches. Graph searches are general, efficient and come with rigorous theoretical analysis. They typically do not apply however to solving problems that involve uncertainty (i.e., Markov Decision Processes). In these projects, we study how various planning under uncertainty problems can be decomposed into a series of graph searches and thus solved efficiently and with solid theory.

Here are few examples of our past work:

  • Probabilistic planning with Clear Preferences. For example, PPCP (Artificial Intelligence Journal '09).
  • Planning for Markov Decision Processes with Sparse Stochasticity. For example, MCP (NIPS '04).

Applications:

Robot-helicopter coordination with MCP. Path Clearance with 5 UAV scouts using PPCP. movie.