Planning Visualizations
Christopher H. Lin and Jonathan Bragg
Abstract
We create a tool for visualizing Markov Decision Processes (MDPs) that assists AI and planning researchers in understanding them. MDPs are a formalism for modeling planning problems. They, as well as their solutions can be complex. Our visualization tool helps researchers to diagnose problems in their model, understand optimal policies, and devise improved algorithms for planning. Using a standard benchmark MDP from the literature, we show that reseachers can answer questions like "Why did my policy choose this action instead of that action?" and "What are the possible values of the states that I might transition to after taking this action?"
Link to paper
Link to poster
Running Instructions
Visit our visualization page here and go!