Autonomous exploration of unknown environments

Thesis defense of Joshua Ott
The capability to autonomously explore unknown environments is becoming increasingly essential as autonomous systems are deployed in diverse applications ranging from planetary exploration, to financial markets, to search and rescue operations. Efficient and safe exploration requires decision-making strategies capable of quantifying uncertainty, optimizing resource use, and maximizing information gain. This thesis addresses these challenges through various approaches to autonomous exploration.
The first part of this defense will focus on the informative path planning problem, including its formulation, measures of informativeness, and the methods we use to solve it. We will also address the adaptive and multimodal variant of the problem, which involves using multiple sensors with varying accuracies and costs. We present various solution methods, ranging from mixed integer programs to sequential tree search-based techniques. We show that our methods are not just limited to the exploration of physical environments but can be extended to the exploration of any abstract parameter space. Specifically, we demonstrate that informative input design for system identification is just a different flavor of an informative path planning problem. We validate our methods through simulations and large-scale field tests, demonstrating their effectiveness in diverse applications like planetary exploration, aircraft flight test, and search and rescue scenarios.