Prior work in cooperative localization for swarms assumed that the basic robots had access to the range and bearing measurements of the advanced robots [2]. With this information, they were able to update the basic robot's heading. We believe that our method could also be able to correct robot heading if the advanced robot estimated the direction of the velocity of the basic robot using multiple successive measurements. Since the basic robot is always moving in the direction of its heading, this would provide a heading estimate of the basic robot.
Another technique to improve the performance of scenarios where multiple advanced robots communicate (i.e. Scenario 2) might be to share the state estimate of the landmarks across all robots capable of sensing the landmarks. Currently, even when communicating, each robot estimates the landmark positions on its own. The CI update step only updates the position of the robot. If robots shared the landmark information as well, however, the combination of all the measurements of all the robots would most likely improve the landmark mapping, thereby improving localization performance as well.
Correct selection of omega is critical to the success of Cooperative Localization via Covariance Intersection.
Single robots with poor state estimates can harm the performance of all other robots.
Covariance Intersection is more sensitive to inaccurate odometry/measurement covariance than EKF.
Covariance Intersection is a viable strategy to improve state estimates of multiple robots, but given the conditions of our dataset, updates from advanced alone are not enough to accurately localize a robot without any localization sensors of its own on it. With more advanced robots and access to more measurements of basic robots, this statement would need to be re-evaluated.
The authors would like to thank Prof. Shia for fantastic instruction in E205 and for invaluable his guidance with the project.