2D-3D hybrid mapping for path planning in autonomous robots

Computational complexity is one of the critical attributes of robot design. Mapping, a vitally important feature of auto-navigation of robots is one such area where computational complexity is of concern. An appropriate spatial representation of the surroundings is required for efficient path planning. Creation of a 2D map alone for a given environment is not sufficient even though it has the least computational complexity. 3D occupancy grid maps and 3D normal distribution maps need larger space and time. This research proposes an efficient hybrid mapping technique (2D-3D h ) which uses both 2D and 3D sensor data mapping to represent an efficient and compact representation of the environments aiming at reducing the computational complexity. The 2D maps are built from the laser data obtained from the Light Detection and Ranging (LiDAR) sensor and 3D maps are built from the data obtained from the Kinect sensor. The proposed algorithm efficiently organizes the 2D-3D h map in such a way that the mapping switches from 2 to 3D only when an elevation region is detected in the environment. The newly generated map consists of both 2D and 3D data, with optimized mapping time and storage capacity requirements. Simulation results indicate that the proposed hybrid mapping can save mapping time up to 46.7% and storage space up to 98.5%. Real time experiments show that 2D-3Dh mapping can save more than 33% of memory space and requires less than half the time compared to that of full 3D mapping.