Symmetric environments pose a potential risk to mobile robots in estimating their pose correctly, especially after a global localization procedure or displacement by an external force (i.e. a kidnapped robot situation). This work presents a system for mobile robots which minimizes the risk of an incorrect pose estimate by using a novel room-awareness module. The module which mimics a human-like belief in the current pose and triggers a so-called spontaneous reorientation in order to solve pose ambiguities in rotationally symmetric environments.
Unlike classical approaches, the robot is able to use the symmetric properties of an environment to, firstly, detect an incorrect pose estimate, and, secondly, to maintain local pose information after the system has detected an incorrect pose.
This is possible because the room-awareness module evaluates the current pose, independent of the robot's self-localization module, based on an online trained model of the visual background which is composed of a spatial colour-histograms features.
The evaluation result is fed into the robot's behaviour controller, which uses a novel interface for the robot's particle-filter-based self-localization to selectively move particle clusters between pose ambiguities in rotationally symmetric environments.
In addition, the room-awareness module is able to directly support the self-localization module by involving the visual background in particle generation in order to prevent particle injection on pose ambiguities.
Tests on a humanoid robot within simulated and real RoboCup Standard Platform League environments demonstrate, on the one hand, the specific challenges to self-localization which generally occur in other robotic set-ups, and on the other hand, the performance gain in pose estimation resulting from the approach presented here.