Complementary integration framework for localization and recognition of a humanoid robot based on task-oriented frequency and accuracy requirements

A robot system that can process environmental measurements and motion planning during locomotion is necessary to continuously perform various tasks. To achieve such a system, which we call the Perception-during-Traversing Model, the accuracy of environmental recognition must be improved and computational costs must be reduced; these are tradeoff relationships. In this paper, we propose a construction framework for a humanoid robot to solve the trade-off problems and achieve the Perception-during-Traversing Model system. The key idea of the proposed framework is subdividing and re-integrating the localization and recognition processes in a complementary manner based on task-oriented frequency and accuracy requirements. Moreover, we apply our framework to the humanoid robot JAXON, and demonstrate that it can execute various tasks continuously by the Perception-during-Traversing Model. The most important contribution of our framework is enabling the humanoid robot to localize itself accurately and measure the environment densely enough to execute tasks using its on-board computers; this provides a practical solution to the trade-off between recognition quality and computational costs.