A Task-Driven Perspective on Status and Development of Machine Map

With the continuous promotion of national robot development strategy, machine map, which guarantees robot cognition and learning, has become a new direction of map science research in recent years. To address the lack of theoretical research on machine map, we summarize the key technologies and research status of task-driven machine maps based on the integrated sensing, mapping and decision-making mechanism. Data acquisition and processing are the basic support for map model construction and task application. And the state of robot platform, feature extraction, semantic segmentation and multi-sensor fusion are summarized. Map model construction is the top and bottom. The commonly used map model architecture and its characteristics are analyzed to describe how to build a resilient, robust, and reliable map building system in different task contexts. Task application is the central expression of the functionality of machine maps. The task applications are the concentrated manifestation of the functionalities of machine maps. And the current status of research on typical applications such as path planning, target detection, knowledge representation and reasoning is introduced and summarized. The problems and future development directions of machine map are discussed.