Nothing has happened in the last couple of weeks, unless you count Sagar and my attempts to document packages from our 2013 GitHub repository. In the same vein, Sagar uploaded all the bags we got during the competition to Dropbox.
We’ve also made a couple of Google docs to gauge and gather interest within RAS for IGVC–both of which were posted in the RAS Facebook group a couple of days ago. So far only Chris Davis, Rachel, Sagar, and Joshua James have put down their interests.
I’ve been thinking about a hypothetical redesign of our current software system. This would only be a reorganization of data flow, packages, and nodes–not any changes to implementation. See the diagram below. Here are the main ideas:
1. The vision nodes should be compressed into one node. If needed, stages in the vision pipeline–along with the threshold/Hough line UI tweaker–would be displayed with OpenCV windows.
2. The Extended Kalman Filter (EKF) should be a service that does a prediction instead of publishing predictions to a topic at 10 Hz.
3. Unnecessary nodes should be removed. Having an entire node dedicated to choosing whether or not to take the VN 200’s INS or GPS topic is unnecessary, and we never used the INS data anyway. The vel_cmd_filter node is also unnecessary–it does no filtering.
4. Scans should be matched with EKF state data to be sure that actions are being made with data all from the same moment in time.
5. Not shown on the diagram: GPS x/y waypoints, red/white thresholds, and Hough line parameters should be specified as ROS parameters in launch files rather than needing to be read from text files.
6. Names for topics, nodes, and packages should be consistent and simple.