Significant Risks
- Roomba odometry data
- The Roomba provides very unreliable odom data. This is backed by many complaints found online about its unreliability.
- This poses an issue for other systems that rely on odom, such as navigation, and gmapping (an alternative SLAM engine that produced worse results when compared to hector SLAM).
- During navigation to a target goal, the Roomba would just spin in a circle.
- We plan on using laser scans to replace odometry data.
- Transparent Objects
- Transparent objects, such as windows, produce unreliable maps. To create better quality maps, we have blocked the bottoms of windows that are inline with our lidar system.
- We are not sure how this will affect localization, further testing is necessary to see if windows need to be blocked during operations.
Design Changes
- Due to the aforementioned issues with Roomba odometer data, we have decided to use laser scans to produce odom data.
Collaborated on Navigation stack bring up.
- We were able to produce odom data by using the laser_scan_matcher package for ROS.
- We have successfully setup navigation on the Xavier NX. We were able to produce a global path to the target, but were unable to get the Roomba to follow the path (it just turned in a circle).