Informative Path Planning to Estimate Quantiles for Environmental Analysis

Scientists interested in studying natural phenomena often take physical specimens from locations the environment for later analysis. These analysis are typically specified by expert heuristics. Instead, we propose to choose scientific using a robot perform an informative path planning survey. The survey results list of that correspond quantile values phenomenon interest. We develop planner novel objective functions improve estimates over time and approach find which values. test our four different environments previously collected aquatic data validate it field trial. Our proposed estimate quantiles has 10.2% mean reduction median error when compared baseline attempts maximize spatial coverage. Additionally, localizing these environment, see 15.7% cross-entropy with loss function baseline.