My team had several work sessions to combine the different subsystems on a basic level. We decided to ditch Cartographer and ROS SLAM for both complexity and compatibility issues, relying instead on an open source Python package called BreezySLAM.

This week, I installed BreezySLAM, ran it through a demo dataset, and extracted the internal data structure representing the SLAM map. I wrote a compression function to reduce the image into a grid heatmap which we plan to do the actual path planning on, since a full-resolution image would be too granular for our hardware. I also wrote a rudimentary navigation planner, which at the moment, when given a “current robot position”, finds an unexplored area within a certain distance.

Next week, I plan to make the navigation much smarter. I plan to implement a search that expands radially outward, in order to prioritize closer points. I also plan to come up with some standard for “exploredness” that is worthy of travel, relative to the other grid squares. Once these are implemented, I plan to write the actual shortest-path algorithm that determines the travel path, which may in turn affect the original search, since it may find a grid on the other side of a wall.

Overall, we are probably slightly behind given that we had several delays with setup and ramp-up for our environments. However, all 3 of us have clearly communicated and divided our work, and every member is making progress in their respective area. We are also regularly having joint work sessions to integrate and decide on next steps.