Cooperative Vision for Swarm Navigation

The efficiency of Mars exploration can be improved by using a swarm of small independent rovers. By using cameras, a rover can estimate its egomotion and build a map of the environment simultaneously. However, the error accumulation in the tracking strongly limits the accuracy of the positioning and the mapping, except if loop closures can be established in the exploring trajectory. We solved the challenges posed by the use of monocular cameras on small rovers and the lack of loop closures by a sensor fusion method based on images and ranges to a single anchor point, e.g., the landing site. Additionally, ranges can be determined between each pair of rovers in a swarm. We proposed a relative pose and scale estimation with low communication requirement and a swarm navigation solution based on that.