This post deals with the Octomaps library which deals with creation of an efficient way of storing, querying and include uncertainty in measurements. Octomaps are data structures t represent 3d space. There are many approaches for modelling 3D space in the robotics domain by creating voxels.

Octomaps differ from them as it does not suffer from space constraints and the problems such as inability to model unknown and free space. Octomaps can model uncertainty in obstacles in the occupancy grid by incorporating probabilistic updates of the occupancy maps. Octomaps can be used to represent voxels at different resolutions. The datastructure used for their implementation are octrees.

Every Octree has eight children. Basically the root of the tree is the area that you are modelling. This region can be unknown initially as the restriction here is the size that you want your octree to be. If the size of the Octree should be small then a dynamically exploring a bigger area would yield leaves with coarser resolutions and vice versa. To include free space in the modelling Octrees explicitly include two types of node: Occupied and Free. Unknown nodes are not generated. Each root is a volume in real world. The children are obtained by dividing the root volume into eight equal parts (voxels) something similar to the four quadrants that we normally use in 2 dimensions. The occupancy values are updated using some distance measurement sensor. The voxels containing the endpoints of the sensor readings are marked as occupied in a discrete map and updated according to some predefined formula in a probabilistic occupancy grid. The region on the ray between the sensor and the endpoint is marked as free and the octree is updated accordingly.  
Currently I am looking at ways to represent the Lidar scans that would be acquired using Octomaps and get a 3D map for the environment. Octomaps are purely for generation of maps. Octomaps are however flexible as they are available as a opensource C++ library which can be included in any C++ project. I am also looking to study the Hector-SLAM package for ROS which is restricted to ROS from what I know. Hector-SLAM though includes mapping, localization which is highly desirable if you want to get started with your planning projects. Currently I am looking to study both of them and try to implement some examples