Consistent Distributed Invariant EKF SLAM with Point and Object Features

Joint-CI is a distributed fusion method for multi-robot right-invariant EKF SLAM with point features, object features, or both. It uses a neighboring robot’s posterior and an inter-robot observation to construct a joint estimate in the receiving robot’s error coordinates, then applies partial-state covariance intersection. The paper reports preserved global rotation and translation unobservability and first-order covariance consistency; simulations and a dataset-derived experiment show improved accuracy with mildly conservative uncertainty.