Adaptive informative path planning (AIPP) is important to many robotics applications, enabling mobile robots to efficiently collect useful data about initially unknown environments. In addition, learning-based methods are increasingly used in robotics to enhance adaptability, versatility, and robustness across diverse and complex tasks. Our survey explores research on applying robotic learning to AIPP, bridging the gap between these two research fields. We begin by providing a unified mathematical problem definition for general AIPP problems. Next, we establish two complementary taxonomies of current work from the perspectives of (i) learning algorithms and (ii) robotic applications. We explore synergies, recent trends, and highlight the benefits of learning-based methods in AIPP frameworks. Finally, we discuss key challenges and promising future directions to enable more generally applicable and robust robotic data-gathering systems through learning. We provide a comprehensive catalog of papers reviewed in our survey, including publicly available repositories, to facilitate future studies in the field.

A common solution criterion for partially observable Markov decision processes (POMDPs) is to maximize the expected sum of exponentially discounted rewards, for which a variety of approximate methods have been proposed. Those that plan in the belief space typically provide tighter performance guarantees, but those that plan over the state space (e.g., QMDP and FIB) often require much less memory and computation. This paper presents an encouraging result that shows that reducing the discount factor while planning in the state space can actually improve performance significantly when evaluated on the original problem. This phenomenon is confirmed by both a theoretical analysis as well as a series of empirical studies on benchmark problems. As predicted by the theory and confirmed empirically, the phenomenon is most prominent when the observation model is noisy or rewards are sparse.