This paper addresses the implementation of an integrity monitor as part of a factor-graph-based absolute and relative position estimator for swarm navigation. Applications that use small Unmanned Aerial Vehicles (sUAV) are increasing in demand and complexity. Using multiple possibly dissimilarly equipped sUAVs to perform tasks such as infrastructure inspection or mapping may not only significantly reduce the time required to complete a task but also reduce the risk of threats to population and property due to the reduced complexity and weight, increased reliability, longer endurance of the smaller platforms, and, most importantly, the ability of the swarm to increase safety by exploiting the increase and reliability of knowledge in distributed cognition and swarm-wide collaboration.
In previous work, the authors introduced the use of factor graph optimization (FGO) for decentralized sUAV position estimation. In the proposed method, each of the swarm members gathered as much observations as possible either directly from its sensors (e.g., range-radios, GNSS, inertial, laser-range scanner, camera), or indirectly from other sUAVs via the swarm network (Figure 1). These observations are then used to setup a factor graph consisting of both measurement and motion model residuals. Finally, this factor graph is optimized to obtain the sUAV position estimate (either absolute or relative). Flight tests and simulation result have shown good estimation accuracy performance in partial GNSS-denied environments when the gathered observations provide sufficient constraints for the algorithm to converge.
In this paper, the authors address the integrity aspects of the proposed method. In particular, the detection of faults (off-nominal conditions) in the measurements (similarly to receiver autonomous integrity monitoring), a method to assess the observability of the estimator, and the selection of the best possible subset of observations if it is overdetermined. The paper will also include some preliminary flight test results with programmed sensor failures.
