Memory Grid Mapping: An Active Map Learning Approach for Autonomous Robots in Unknown Environments

摘要

This paper proposes memory grid mapping, a new approach to active map learning for autonomous robots exploring in unknown indoor environments. Memory grid mapping makes use of a map model, and methods for updating maps, exploration, and map postprocessing and adopts a grid-based representation and a frequency value to measure the confidence that a cell is occupied by an obstacle. The fast map updating and path planning (i.e. the exploration method) make our approach a candidate for real-time implementation on mobile robots. The exploration method has focused on fast path planning (benefit from planning in a fixed regional range) rather than optimal path (benefit from global search). The map postprocessing method is effective to increase the accuracy of learned map. The general framework of map postprocessing involves a threshold operation, a template operation and an insert operation. The approach has no any assumption of environmental complexity and obstacle shape or size. The experimental results are demonstrated by simulated tests using a Pioneer robot with eight forward sonar sensors.