Robust and Efficient Slam Via Compressed H ∞ Filtering

Abstract Simultaneous localization and mapping ( SLAM ) is an important and challenging task for the operation of autonomous mobile robots in which both the pose of the robot and features of the environment need to be estimated at the same time. In particular, it is desirable to achieve robustness and efficiency in the SLAM implementation. Robots in unknown environment are likely to be subject to modeling errors which cannot be easily characterized in terms of statistical properties. To mitigate the effect of uncertainties and disturbances, robust filters such as the H ∞ filter can be employed. However, robust filters are complex to implement, demanding a significant amount of computational resources. This study proposes a compressed H ∞ filter to solve the robust SLAM problem in which robot dynamics are subject to uncertainties and measurements are subject to bounded‐but‐unknown disturbances. To achieve an efficient implementation, the state is partitioned into active and inactive states where the latter refers to state variables which are invariant and independent of the measurement at the epoch. With such a partitioning, the active state represents the robot pose and locations of landmarks inside a certain area surrounding the robot. The computational load is reduced since only active state needs to be estimated within a time segment. An update scheme is proposed to refine the whole state at the end of the time segment. Moreover, the relative landmark representation which results in a small cross‐correlation between active state and inactive state is employed to reduce the errors. As a result, both efficiency and robustness can be achieved. Simulations reveal that the results obtained from the proposed compressed H ∞ filter, which has lower computational complexity, are very close to those from the full order H ∞ filter. Further, the compressed H ∞ filter is more robust than EKF s and FastSLAM .