This project involves the development of sampling policies to map the spatial variability of certain environmental factors. The robot is given a finite set of locations and the goal is to sample a subset of the locations to capture the variability. Depending on the type of information given to the robot ahead of time, the strategy used to sample varies. For example, if a previous imperfect map of the environment is given, one can sample at locations where the variance is maximum to reduce uncertainty. This falls under passive/offline sampling. In another case, there may be no prior information given to the robot. Instead, it gains information about the environment as it processes samples sequentially, known as online/active learning. We are interested in developing efficient and optimal algorithms for both types of problems.
