POMDP Library Optimizing Over Exploration and Exploitation in Robotic Localization, Mapping, and Planning

Abstract

Localization, mapping, and planning are critical in autonomous robots operating in uncertain environments and in continuous and discrete domains. High-quality probabilistic models for a complex robot depend heavily on details from its environment, involving multiple parameters. However, there is a lack of accurate probabilistic models for existing robots that can handle reasonably the challenges posed by real applications. For most robots, actions are highly non-deterministic. Furthermore, there is a lack of general software packages applicable to new scenarios. Specifically, we propose a POMDP library for optimal planning and localization given new available models, and dedicated to optimize over exploration and exploitation tradeoffs.