Efficient Distributed State Estimation for Swarm Robotics

Astrobotic and CMU will develop a real time distributed localization method named DALEC that combines local visual-inertial odometry with Ultra-wideband (UWB) range measurements between rovers to improve each vehiclersquo;s localization and provide each with information about the othersrsquo; location. Because of a lack of global constraints, visual inertial odometry (VIO) will drift over time in four degrees of freedom. Range information between rovers will provide additional constraints that limit drift. To perform distributed localization, each rover will estimate its own trajectory in a factor graph of poses and receive additional condensed information from other rovers. Because the range and VIO measurements are not tightly coupled, each of the rovers can navigate on its own, and this method is inherently robust to communication outages. A key advantage of this approach is that any other constraints from sensors such as sun angle, bearing to visually observed rovers, terrain matching, and point features, can easily be added to the factor graph formulation.The proposed technical approach is most closely related to the Decentralized Data Fusion Smoothing and Mapping algorithm (DDF-SAM2, specifically), though DDF-SAM2 assumes landmark (point) feature observations and shares the relative position of these landmarks by default. It can also be extended to share relative positions and orientations of objects rather than just the positions of landmarks.nbsp;Our problem is different in that only range measurements are available from the UWB sensor, but these are measurements with error and drift models that are much simpler and do not require solving a data association or loop closing problem.A simulation environment will be developednbsp;to assist in the development and testing of the DALEC SLAM algorithm. Thenbsp;simulation will produce simulated IMU, UWB, and imagery data from 4ndash;15 rovers operating simultaneously. The simulation will then be used to test and tune the DALEC algorithm.