NAMOUnc: Navigation among movable obstacles with decision making on uncertainty interval

Navigation among movable obstacles~(NAMO) is a fundamental task in various applications. Existing NAMO solutions usually assume that the robot operates in an ideal world with perfect observation and actions or a full knowledge on the environment. These assumptions limit their applicability on real robots and may result in risky actions. We propose a method~(NAMOUnc) with consideration of multiple uncertainties, including observation noise, action failure, prediction uncertainty and uncertainty caused by partial observability. Based on the estimated uncertainties, the robot makes decision by comparing the time cost interval to reach the goal, then achieves the joint optimization on the time cost and success rate. We evaluate the proposed algorithm in both simulated and real environments and compare it with latest NAMO frameworks.