Decide on a mapping framework

[steph1111]

We plan to use D-Star Lite to navigate the robot from one point to another in 3D space dynamically avoiding obstacles. D-Star lite traverses through a updating graph to find the shortest path. We therefore need a way to representing 3D space as a graph-like structure. This should be done by creating a voxel grid of the observable space--3D cubes with dimensions the longest robot measurement which are either occupied or free. Ideally, we would like to map the entire course using voxels so we can complete the go home task from any location. This means we need some sort of SLAM system.

Finding a mapping framework which does exactly what we want has been difficult, and making one ourselves is complicated due to cv loop closure. As a result the software has been at a stand-still. I am making a hard deadline of 6/18 to choose what we are using, and we are choosing between two options.

Octomap

• The issue was converting a binary octree to a traversable graph. We can use conversions.h, cpp builds in the repo on the cpp branch.
• We would also need some sort of SLAM for loop closure.

NvBlox

• isaac_ros_nvblox
• ROS Topics and Services ~/dynamic_occupancy_layer looks interesting
• Looks like you can set the voxel size with parameters (see)
• If we just intercept it before it gets reduced to 2d this could be good
• Must run on the Jetson
• Example launch: nvblox.launch.py
• It also interfaces with zed ZED Camera Examples